File size: 138,551 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T06:07:34.110950Z"
},
"title": "Multi-Emotion Classification for Song Lyrics",
"authors": [
{
"first": "Darren",
"middle": [],
"last": "Edmonds",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "ICS University of California",
"location": {
"settlement": "Irvine"
}
},
"email": "dedmond1@uci.edu"
},
{
"first": "Jo\u00e3o",
"middle": [],
"last": "Sedoc",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "New York University",
"location": {}
},
"email": "jsedoc@stern.nyu.edu"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Song lyrics convey a multitude of emotions to the listener and powerfully portray the emotional state of the writer or singer. This paper examines a variety of modeling approaches to the multi-emotion classification problem for songs. We introduce the Edmonds Dance dataset, a novel emotion-annotated lyrics dataset from the reader's perspective, and annotate the dataset of Mihalcea and Strapparava (2012) at the song level. We find that models trained on relatively small song datasets achieve marginally better performance than BERT (Devlin et al., 2019) finetuned on large social media or dialog datasets.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "Song lyrics convey a multitude of emotions to the listener and powerfully portray the emotional state of the writer or singer. This paper examines a variety of modeling approaches to the multi-emotion classification problem for songs. We introduce the Edmonds Dance dataset, a novel emotion-annotated lyrics dataset from the reader's perspective, and annotate the dataset of Mihalcea and Strapparava (2012) at the song level. We find that models trained on relatively small song datasets achieve marginally better performance than BERT (Devlin et al., 2019) finetuned on large social media or dialog datasets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Text-based sentiment analysis has become increasingly popular in recent years, in part due to its numerous applications in fields such as marketing, politics, and psychology (Rambocas and Pacheco, 2018; Haselmayer and Jenny, 2017; Provoost et al., 2019) . However, the vast majority of sentiment analysis models are built to identify net positive or negative sentiment rather than more complex, ambiguous emotions such as anticipation, surprise, or nostalgia (Jongeling et al., 2017) . As a result, current models usually fail to portray the coexistence of multiple emotions within a text sample, resulting in limited characterization of a human's true emotions. Songs are often created to elicit complex emotional responses from listeners, and thus are an interesting area of study to understand nuanced emotions .",
"cite_spans": [
{
"start": 174,
"end": 202,
"text": "(Rambocas and Pacheco, 2018;",
"ref_id": "BIBREF37"
},
{
"start": 203,
"end": 230,
"text": "Haselmayer and Jenny, 2017;",
"ref_id": "BIBREF16"
},
{
"start": 231,
"end": 253,
"text": "Provoost et al., 2019)",
"ref_id": "BIBREF35"
},
{
"start": 459,
"end": 483,
"text": "(Jongeling et al., 2017)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This paper examines a variety of approaches to address the multi-emotion classification problem. We aim to build an emotion classification model that can detect the presence of multiple emotions in song lyrics with comparable accuracy to the typical inter-annotator agreement for textbased sentiment analysis (70-90%) (Diakopoulos and Shamma, 2010; Bobicev and Sokolova, 2017; Takala et al., 2014) . Building such a model is especially challenging in practice as there often exists considerable disagreement regarding the perception and interpretation of the emotions of a song or ambiguity within the song itself (Kim et al., 2010) .",
"cite_spans": [
{
"start": 318,
"end": 348,
"text": "(Diakopoulos and Shamma, 2010;",
"ref_id": "BIBREF10"
},
{
"start": 349,
"end": 376,
"text": "Bobicev and Sokolova, 2017;",
"ref_id": "BIBREF4"
},
{
"start": 377,
"end": 397,
"text": "Takala et al., 2014)",
"ref_id": "BIBREF44"
},
{
"start": 614,
"end": 632,
"text": "(Kim et al., 2010)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There exist a variety of high-quality text datasets for emotion classification, from social media datasets such as CBET (Shahraki, 2015) and TEC (Mohammad, 2012) to large dialog corpora such as the DailyDialog dataset (Li et al., 2017) . However, there remains a lack of comparable emotionannotated song lyric datasets, and existing lyrical datasets are often annotated for valence-arousal affect rather than distinct emotions (\u00c7ano and Morisio, 2017) . Consequently, we introduce the Edmonds Dance Dataset 1 , a novel lyrical dataset that was crowdsourced through Amazon Mechanical Turk. Our dataset consists of scalar annotations for the 8 core emotions presented by Plutchik (2001) , with annotations collected at the song level and from the reader's perspective.",
"cite_spans": [
{
"start": 120,
"end": 136,
"text": "(Shahraki, 2015)",
"ref_id": "BIBREF40"
},
{
"start": 145,
"end": 161,
"text": "(Mohammad, 2012)",
"ref_id": "BIBREF28"
},
{
"start": 218,
"end": 235,
"text": "(Li et al., 2017)",
"ref_id": "BIBREF24"
},
{
"start": 427,
"end": 451,
"text": "(\u00c7ano and Morisio, 2017)",
"ref_id": "BIBREF6"
},
{
"start": 669,
"end": 684,
"text": "Plutchik (2001)",
"ref_id": "BIBREF33"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We find that BERT models trained on out-ofdomain data do not generalize well to song lyrics and have lower F1 scores than Naive Bayes classifiers for emotions such as disgust and fear. However, BERT models trained on small lyrical datasets achieve marginally better performance, despite indomain datasets being orders of magnitude smaller than their counterparts. We also find that surprise has significantly lower inter-annotator agreement and test accuracy than other core emotions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "A multitude of models and techniques have been explored for song emotion classification. Both He et al. (2008) and Wang et al. (2011) found that fea-ture extraction from lyrics improves emotion classification performance. Researchers have trained Naive Bayes, HMM, SVM, clustering, and Random Forest models on lyrical and sometimes audio features to predict emotion in songs (Hu et al., 2009; Kim and Kwon, 2011; Jamdar et al., 2015; Rachman et al., 2018) . Deep learning frameworks have also been widely utilized for song emotion classification, ranging from CNNs and LSTMs (Delbouys et al., 2018; Abdillah et al., 2020) to transformer-based models such as BERT and ELMo (Parisi et al., 2019; Liu and Tan, 2020) .",
"cite_spans": [
{
"start": 94,
"end": 110,
"text": "He et al. (2008)",
"ref_id": "BIBREF17"
},
{
"start": 115,
"end": 133,
"text": "Wang et al. (2011)",
"ref_id": "BIBREF45"
},
{
"start": 375,
"end": 392,
"text": "(Hu et al., 2009;",
"ref_id": "BIBREF19"
},
{
"start": 393,
"end": 412,
"text": "Kim and Kwon, 2011;",
"ref_id": "BIBREF22"
},
{
"start": 413,
"end": 433,
"text": "Jamdar et al., 2015;",
"ref_id": "BIBREF20"
},
{
"start": 434,
"end": 455,
"text": "Rachman et al., 2018)",
"ref_id": "BIBREF36"
},
{
"start": 575,
"end": 598,
"text": "(Delbouys et al., 2018;",
"ref_id": "BIBREF8"
},
{
"start": 599,
"end": 621,
"text": "Abdillah et al., 2020)",
"ref_id": "BIBREF1"
},
{
"start": 672,
"end": 693,
"text": "(Parisi et al., 2019;",
"ref_id": "BIBREF32"
},
{
"start": 694,
"end": 712,
"text": "Liu and Tan, 2020)",
"ref_id": "BIBREF25"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Multiple researchers have taken a multi-modal approach to emotion prediction. , introduced a novel corpus of both music and lyrics, and achieved promising results when using both musical and lyrical representations of songs in emotion classification. Similarly, Yang et al. (2008) found an increase in 4-class emotion prediction accuracy from 46.6 to 57.1 percent when incorporating lyrics into models trained on audio.",
"cite_spans": [
{
"start": 262,
"end": 280,
"text": "Yang et al. (2008)",
"ref_id": "BIBREF47"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "However, audio data can lead to problematic bias in emotion classification. Susino and Schubert (2019b) explored the presence of emotion stereotyping in certain genres, and found that heavy metal and hip-hop music were perceived to have more negative emotions than pop music with matched lyrics. Susino and Schubert (2019a) also found that emotional responses to an audio sample of a song could be predicted by stereotypes of the culture with which the song's genre was associated. Additionally, Fried (1999) found that violent lyrical passages were seen to be significantly more negative when represented as rap songs rather than country songs. Dunbar et al. (2016) validated Fried's findings through multiple studies in which participants believed that identical lyrics were more offensive when portrayed as rap rather than country music.",
"cite_spans": [
{
"start": 76,
"end": 103,
"text": "Susino and Schubert (2019b)",
"ref_id": "BIBREF43"
},
{
"start": 296,
"end": 323,
"text": "Susino and Schubert (2019a)",
"ref_id": "BIBREF42"
},
{
"start": 482,
"end": 508,
"text": "Additionally, Fried (1999)",
"ref_id": null
},
{
"start": 646,
"end": 666,
"text": "Dunbar et al. (2016)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Lyrics are paramount for the accurate prediction of emotion in music. Yang and Lee (2009) transformed song lyrics into psychological feature vectors using a content analysis package and concluded that song lyrics alone can be used to generate promising, human-comprehensible classification models. Hu et al. (2009) found that audio features did not always outperform lyric features for mood prediction, and that combining lyric and audio features does not necessarily improve mood prediction over simply training on lyrics features. In later research, Hu and Downie (2010) found that lyrics features significantly outperformed au-dio features in 7 of 18 mood categories, while audio features outperformed lyrical features in only one.",
"cite_spans": [
{
"start": 70,
"end": 89,
"text": "Yang and Lee (2009)",
"ref_id": "BIBREF46"
},
{
"start": 298,
"end": 314,
"text": "Hu et al. (2009)",
"ref_id": "BIBREF19"
},
{
"start": 552,
"end": 572,
"text": "Hu and Downie (2010)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Research is split regarding crowdsourced emotion annotation quality; while Mohammad and Bravo-Marquez (2017) achieved strong results through crowdsourcing labels, Hasan et al. (2014) found crowd labels to sometimes not even be in agreement with themselves. Surprise is an emotion that is especially difficult to model (Buechel and Hahn, 2017; Schuff et al., 2017) , less frequent (Oberl\u00e4nder and Klinger, 2018) , and is sometimes divided into positive and negative surprise (Alm et al., 2005) .",
"cite_spans": [
{
"start": 75,
"end": 108,
"text": "Mohammad and Bravo-Marquez (2017)",
"ref_id": "BIBREF29"
},
{
"start": 163,
"end": 182,
"text": "Hasan et al. (2014)",
"ref_id": "BIBREF15"
},
{
"start": 318,
"end": 342,
"text": "(Buechel and Hahn, 2017;",
"ref_id": "BIBREF5"
},
{
"start": 343,
"end": 363,
"text": "Schuff et al., 2017)",
"ref_id": "BIBREF39"
},
{
"start": 396,
"end": 410,
"text": "Klinger, 2018)",
"ref_id": "BIBREF31"
},
{
"start": 474,
"end": 492,
"text": "(Alm et al., 2005)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Lyrics are valuable for song emotion prediction and decent classification models can be generated solely on song lyrics. However, many lyrical datasets for song emotion classification are based on valence-arousal and lack emotions such as surprise or fear, which are important components of mood (Ekman and Friesen, 2003) . In addition, there is a lack of large, high quality datasets capturing complex emotion in music.",
"cite_spans": [
{
"start": 296,
"end": 321,
"text": "(Ekman and Friesen, 2003)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "In-domain Datasets",
"sec_num": "3.1"
},
{
"text": "Consequently, we created the Edmonds Dance dataset, a novel corpus of English song lyrics annotated for emotion from the reader's perspective. By searching a Spotify playlist consisting of 800 songs, both lyrical and instrumental, and collecting available lyrics from LyricFind, Genius, and MusixMatch (Lyr; Gen; Mus), we retrieved lyrics for 524 songs. We then labeled our dataset based on Plutchik's 8 core emotions of Anger, Anticipation, Disgust, Fear, Joy, Sadness, Surprise, and Trust (Plutchik, 2001) . Table 1 depicts a subsection of the Edmonds Dance dataset, while the Appendix has more information on our labeling methods.",
"cite_spans": [
{
"start": 491,
"end": 507,
"text": "(Plutchik, 2001)",
"ref_id": "BIBREF33"
}
],
"ref_spans": [
{
"start": 510,
"end": 517,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "A Novel Lyrics Dataset Annotated for Emotion",
"sec_num": null
},
{
"text": "In addition to the Edmonds Dance dataset, we also reannotated the dataset introduced in Mihalcea and Strapparava (2012), a multimodal corpus of songs that includes scalar annotations of both audio and lyrics for Ekman's six core emotions: Anger, Disgust, Fear, Joy, Sadness, and Surprise (Ekman, 1993) . The original dataset was annotated from the songwriter's perspective and at a line level. We averaged these line-level lyrical annotations to achieve classifications at higher levels, thus gen- Annotation Guidelines To generate reliable annotations, our HIT included detailed annotation instructions. We organized these guidelines into four sections: initial instructions, important notes, definitions, and examples. The initial instructions section provided the annotator with basic task information, stating that he or she will be given a set of song lyrics, and is expected to record the degree to which the lyrics contain eight specific emotions. We also stated that emotions would be rated on a 6-point scale ranging from the complete absence of an emotion to the extreme presence of an emotion.",
"cite_spans": [
{
"start": 288,
"end": 301,
"text": "(Ekman, 1993)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Mihalcea/Strapparava Dataset Reannotation",
"sec_num": null
},
{
"text": "Hip Hop (6.5 %)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Country (1.1 %)",
"sec_num": null
},
{
"text": "Pop (12.9 %) EDM (79.6 %) The important notes section emphasized that English speakers were required for the task, and that completion of all fields was required. The definitions section provided dictionary-level definitions for each of the eight emotions, while the examples section provided two annotated examples, along with general annotation guidelines (see Appendix A.2 for HIT images).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Country (1.1 %)",
"sec_num": null
},
{
"text": "Each HIT contained the same two example songs. Each of the eight emotions was present in at least one of the songs, and emotions evoked by each song were apparent from the lyrics. Our HITs are available upon request.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Country (1.1 %)",
"sec_num": null
},
{
"text": "We evaluated annotator reliability by calculating the average Cohen's Kappa of each annotator against others assigned to the same HIT, and discarding those below the threshold of 0.25. We then analyzed agreement across emotions by calculating Krippendorf's Alpha on the remaining annotators, and examined the agreement between original and Turker annotations using Pearson's correlation coefficient. Table 3 depicts our results, with more details available in the Appendix.",
"cite_spans": [],
"ref_spans": [
{
"start": 400,
"end": 407,
"text": "Table 3",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Error Analysis",
"sec_num": null
},
{
"text": "Surprise had significantly lower inter-annotator agreement than other emotions. Krippendorf's Alpha and Pearson's Correlation values were lowest for Surprise, with significant correlation differences compared to all other emotions except Anticipation. Meanwhile, Joy and Sadness had relatively higher alpha and correlation values, suggesting a hierarchy of difficulty in emotion classification. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error Analysis",
"sec_num": null
},
{
"text": "To confirm the quality of our dataset, we analyzed differences in annotation patterns between included and discarded Turkers. Discarded annotators had lower median completion time across the Edmonds Dance and Mihalcea/Strapparava datasets (p<.005), were more likely to say that they disliked a song (p<.005), and were less likely to say that they were unfamiliar with a song (p<.001). We also found that discarded annotators spent less time than included annotators on labeling songs that they disliked (p<.001). Further details are in the Appendix.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Analysis of Crowd Workers",
"sec_num": null
},
{
"text": "To explore the efficacy of out-of-domain model training, we used the CBET (Shahraki, 2015) , TEC (Mohammad, 2012) , and DailyDialog (Li et al., 2017) To train more robust baseline models, we also created augmented and transformed versions of the datasets; details on this process are available in the Appendix. While no versions of the CBET, TEC, and DailyDialog datasets include music lyrics, they are large enough to train deep models which we hypothesized could accurately predict emotions in smaller, gold-standard test datasets of song lyrics.",
"cite_spans": [
{
"start": 74,
"end": 90,
"text": "(Shahraki, 2015)",
"ref_id": "BIBREF40"
},
{
"start": 97,
"end": 113,
"text": "(Mohammad, 2012)",
"ref_id": "BIBREF28"
},
{
"start": 132,
"end": 149,
"text": "(Li et al., 2017)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Out of Domain Datasets",
"sec_num": "3.2"
},
{
"text": "We chose Naive Bayes as our first baseline emotion classification model due to its widespread applications in text classification and sentiment analysis (Raschka, 2014) . Given its robustness to outliers and ability to deal with imbalanced data (Chen et al., 2004) , a Random Forest baseline model was also implemented. Lastly, we utilized a Most Frequent Sense (MFS) baseline model, given its strong performance in word sense disambiguation tasks and its applications to emotion classification (Preiss et al., 2009) . We trained our Naive Bayes model on bag-of-words features and our Random Forest model on transformed feature vectors which were generated from our textual datasets using the NRC Hashtag Emotion Lexicon (Mohammad and Turney, 2013) ; see Appendix for further details.",
"cite_spans": [
{
"start": 153,
"end": 168,
"text": "(Raschka, 2014)",
"ref_id": "BIBREF38"
},
{
"start": 245,
"end": 264,
"text": "(Chen et al., 2004)",
"ref_id": "BIBREF7"
},
{
"start": 495,
"end": 516,
"text": "(Preiss et al., 2009)",
"ref_id": "BIBREF34"
},
{
"start": 721,
"end": 748,
"text": "(Mohammad and Turney, 2013)",
"ref_id": "BIBREF30"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Model Implementation",
"sec_num": "4"
},
{
"text": "To improve upon emotion classification quality, we also explored more complex models. Due to its ability to generate powerful contextualized word embeddings and its state-of-the-art results in numerous language understanding tasks (Devlin et al., 2019) , the BERT BASE uncased architecture was fine-tuned for multi-emotion classification from the text of song lyrics. BERT BASE consists of 12 Transformer blocks, a hidden size of 768, 12 self-attention heads, and an additional output layer which we used for fine-tuning. 2",
"cite_spans": [
{
"start": 231,
"end": 252,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Model Implementation",
"sec_num": "4"
},
{
"text": "We trained separate BERT models for each emotion on the original and augmented CBET datasets, and tested their performance on the Edmonds Dance and Mihalcea/Strapparava datasets. We then compared these results with those of our baseline Naive Bayes, Random Forest, and Most Frequent Sense models. To compare emotion prediction accuracy across multiple text corpora, we also trained BERT models on the TEC and DailyDialog datasets, and tested them on our lyrical datasets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5"
},
{
"text": "We found that BERT models trained on the CBET, TEC, and DailyDialog datasets did not generalize well to lyrical data. While models for joy and sadness improved upon the performance of baseline classifiers, models for disgust and fear performed worse than our Naive Bayes baseline. Furthermore, data augmentation techniques improved the performance of our baseline Naive Bayes model, but did not significantly increase BERT model accuracy.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5"
},
{
"text": "To compare in-domain model accuracy with our out-of-domain results, we trained and tested BERT models on the Edmonds Dance and Mihalcea/Strapparava datasets, and vice versa. Models trained and tested on lyrical datasets had marginally better accuracy and F1 scores than out-of-domain models for anger, joy, and sadness. Given the much smaller sizes of lyrical datasets compared to their counterparts, as well as the differences in song genre and annotation perspective across lyrical datasets, our findings suggest a significant advantage in using in-domain data to train models for complex emotion classification of songs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5"
},
{
"text": "Finally, all models performed poorly when classifying surprise, and F1 scores for anger, disgust, and fear remained consistently low across models, suggesting a steep hierarchy of difficulty regarding emotion classification. Inter-annotator agreement was much lower for surprise than other emotions, and none of our models were able to accurately predict the presence of surprise in song lyrics. Our work implies that surprise is unique from the perspective of emotion classification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5"
},
{
"text": "Tables 5 and 6 highlight our model results. A complete version of our evaluation results is available in the Appendix. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5"
},
{
"text": "In this paper we explore a variety of approaches to the multi-emotion classification problem for songs. We introduce the Edmonds Dance dataset, a novel lyrical dataset annotated for emotion at the song level and from the reader's perspective. We find that emotion classification of song lyrics using state-of-the-art methods is difficult to accomplish using out-of-domain data; BERT models trained on large corpora of tweets and dialogue do not generalize to lyrical data for emotions other than joy and sadness, and are outperformed by Naive Bayes classifiers on disgust and fear. On the other hand, models trained on song lyrics achieve comparable accuracy to models trained on out-of-domain data, even when lyrical datasets are orders of magnitude smaller than their counterparts, have been aggregated from line to song level, have been annotated from different perspectives, and are composed of different genres of music. Our findings underscore the importance of using in-domain data for song emotion classification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "Our dataset was annotated by 184 Amazon Mechanical Turk crowdworkers. Annotators were paid $0.15 per task or \u223c $6.75 per hour, and reliable annotators (see Appendix A.2) were awarded a bonus of $0.10 per task or \u223c $11.25 per hour.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Ethical Consideration",
"sec_num": "7"
},
{
"text": "The Mihalcea/Strapparava dataset initially consisted of 4976 lines across 100 songs which were annotated using a scale from 0 to 10, with 0 as the absence of emotion and 10 as the highest intensity of emotion. Annotations were based on Ekman's six core emotions: Anger, Disgust, Fear, Joy, Sadness, and Surprise (Ekman, 1993) . As the dataset was annotated at a line level, we averaged emotion annotations on each line to achieve classifications at higher levels. Through averaging, we generated 452 verse-based and 100 song-based annotations. With regards to the Edmonds Dance Dataset, the basis for label selection was provided by Plutchik's Theory of Emotion, which postulates that all emotions are combinations of the 8 core emotions present in our label (Plutchik, 2001) . As a result, the label can lead to additional classification models for emotions which are theorized to be dyads of the core emotions (e.g, PLove = PJoy * PTrust, or PAggressiveness = PAnger * PAnticipation ). Our dataset was initially labeled using an array of size 8; each array index contained a binary value to indicate an emotion's presence. ",
"cite_spans": [
{
"start": 312,
"end": 325,
"text": "(Ekman, 1993)",
"ref_id": "BIBREF12"
},
{
"start": 759,
"end": 775,
"text": "(Plutchik, 2001)",
"ref_id": "BIBREF33"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "A.1 Lyrical Datasets",
"sec_num": null
},
{
"text": "To evaluate the reliability of our Mechanical Turk annotations, we first used Cohen's Kappa to calculate the average inter-annotator agreement of each Turker against others assigned to the same HIT. We then discarded all annotators who failed to meet a threshold of 0.25, and calculated average agreement for each emotion using Krippendorf's Alpha on the remaining annotators. Krippendorf's Alpha values were highest for the emotions of joy, sadness, and trust; additionally, alpha values were relatively consistent across emotions. 31.6% of annotations in Mihalcea and Strapparava's dataset failed to meet the Cohen's Kappa threshold, while 63.2% of annotations in the Edmonds Dance dataset failed to meet the threshold. Our results are summarized in Table A1 , while Figures A1, A2, and A3 depict pictures of our HITs. Next, we calculated the Pearson's correlation coefficient and related p-values between original annotations and the Turker annotations for both the Edmonds Dance and Mihalcea/Strapparava datasets. We were also unable to calculate correlation coefficients for Anticipation or Trust in the Rada dataset as the original dataset did not include annotations for these emotions. These results are summarized in Table A2 .",
"cite_spans": [],
"ref_spans": [
{
"start": 752,
"end": 760,
"text": "Table A1",
"ref_id": "TABREF1"
},
{
"start": 1226,
"end": 1234,
"text": "Table A2",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "A.2 Annotator Error Analysis",
"sec_num": null
},
{
"text": "While the relative strength of Pearson's Correlations across emotions was similar to that of our alpha values, correlation with fear was relatively higher than expected, and correlation with anger and surprise were lower than expected. Finally, we looked at Krippendorf's Alpha Values on an an- Figure A3 : HIT Annotation Format notation group level to better understand whether annotation agreement for specific emotions were consistently similar across songs. Our results, summarized in Table A3 , provide evidence for a hierarchy of difficulty in emotion classification. Joy and Sadness have the most favorable distribution of alpha values with few low item-level alpha scores (<0.2), and greater numbers of medium (0.2-0.6) and high (>0.6) item-level alpha scores. We then created a heat map of item-level Krippendorf's Alphas to explore correlation of interannotator agreement across emotions. Our results, visualized in Figure A4 , reveal that alpha values are only slightly correlated across emotions. This implies that classification difficulty of a specific emotion varies depending on the song being an- Figure A5 : Annotation Completion Time by Quantile notated; indeed, the only emotions that have an inter-annotator agreement correlation above 0.4 are Anger/Disgust and Anger/Anticipation. We can also see that only joy has a moderate correlation with overall agreement across emotions, implying that songs with annotation agreement regarding joy may be easier to classify overall, but songs with annotation agreement regarding other emotions may not necessarily be easier to annotate. Consequently, the claim of a consistent hierarchy of difficulty is somewhat undermined and instead it seems that classification difficulty of a specific emotion varies depending on the song being annotated.",
"cite_spans": [],
"ref_spans": [
{
"start": 295,
"end": 304,
"text": "Figure A3",
"ref_id": null
},
{
"start": 489,
"end": 497,
"text": "Table A3",
"ref_id": "TABREF4"
},
{
"start": 926,
"end": 935,
"text": "Figure A4",
"ref_id": null
},
{
"start": 1114,
"end": 1123,
"text": "Figure A5",
"ref_id": null
}
],
"eq_spans": [],
"section": "A.2 Annotator Error Analysis",
"sec_num": null
},
{
"text": "We analyzed the completion time of annotations across good and bad annotators for the Edmonds Dance and Mihalcea/Strapparava datasets, summarized in Figure A5 . We can see that the distributions of completion times were very similar for bad annotators, while the distributions for good annotators were skewed upwards at higher deciles. In addition, the median completion time for good annotators was 31 seconds greater than the median completion time for bad annotators, and the mean completion time for good annotators was 37 seconds greater than that of bad annotators.",
"cite_spans": [],
"ref_spans": [
{
"start": 149,
"end": 158,
"text": "Figure A5",
"ref_id": null
}
],
"eq_spans": [],
"section": "A.3 Analysis of Crowd Workers",
"sec_num": null
},
{
"text": "Next, we looked at differences between good and bad annotator groups regarding annotator enjoyment and familiarity of labeled songs. We found that bad annotators were more likely than good annotators to say that they were familiar with a song (p<.00001), or that they disliked a song (p<.005). Bad annotators also spent significantly less time than good annotators on labeling songs that they said they disliked (p<0.0001). These results are summarized in Figures A6 and A7 . ",
"cite_spans": [],
"ref_spans": [
{
"start": 456,
"end": 473,
"text": "Figures A6 and A7",
"ref_id": null
}
],
"eq_spans": [],
"section": "A.3 Analysis of Crowd Workers",
"sec_num": null
},
{
"text": "To address misclassification of the minority class, we implemented oversampling techniques so classes would be more balanced. For each emotion in the CBET dataset, we added copies of tweets suggesting the presence of the emotion such that the new number of tweets with presence of emotion was between 40-60 percent of the total number of tweets. We then trained two BERT models for each emotion, one using the augmented CBET dataset and the other using the original. To confirm the quality of the original CBET dataset, we also trained and tested BERT models on subsets of CBET data that were randomly sampled without replacement. These results are shown in Table A4 .",
"cite_spans": [],
"ref_spans": [
{
"start": 658,
"end": 666,
"text": "Table A4",
"ref_id": "TABREF6"
}
],
"eq_spans": [],
"section": "A.4.1 Data Augmentation and Transformation",
"sec_num": null
},
{
"text": "To explore another approach to the multiemotion classification problem, lyrical data was then transformed into a feature vector of length 9 using the NRC Hashtag Emotion Lexicon, which contains binary indicators regarding the presence or absence of Plutchik's 8 core emotions in 14182 common English words (Mohammad and Turney, 2013) . This occurred by iterating through a song's lyrics, counting each word present in the NRC Emotion Lexicon as well as its emotional classification, and storing this information in the feature vector. For example, the feature vector [5, 10, 1, 9, 4, 2, 2, 3, 28] would correspond to a song's lyrics that contained 28 words (not necessarily all distinct) which were present in the NRC Emotion Lexicon. Of these words, 5 were associated with joy, 10 with trust, 1 with fear, etc. This transformed dataset was generated for the purpose of exploring Random Forest methods for song emotion classification.",
"cite_spans": [
{
"start": 306,
"end": 333,
"text": "(Mohammad and Turney, 2013)",
"ref_id": "BIBREF30"
},
{
"start": 567,
"end": 570,
"text": "[5,",
"ref_id": null
},
{
"start": 571,
"end": 574,
"text": "10,",
"ref_id": null
},
{
"start": 575,
"end": 577,
"text": "1,",
"ref_id": null
},
{
"start": 578,
"end": 580,
"text": "9,",
"ref_id": null
},
{
"start": 581,
"end": 583,
"text": "4,",
"ref_id": null
},
{
"start": 584,
"end": 586,
"text": "2,",
"ref_id": null
},
{
"start": 587,
"end": 589,
"text": "2,",
"ref_id": null
},
{
"start": 590,
"end": 592,
"text": "3,",
"ref_id": null
},
{
"start": 593,
"end": 596,
"text": "28]",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "A.4.1 Data Augmentation and Transformation",
"sec_num": null
},
{
"text": "To gauge the quality of the CBET dataset, we first calculated the accuracies of BERT models trained and tested on randomly ordered subsets of CBET data, with an 80/20 train/test split. Emotion classification accuracies of these models were at least 90%, confirming the quality of the dataset. Next, we trained BERT models on the full CBET datasets, and evaluated them on the verse-based variation of Mihalcea and Strapparava's dataset, as well as the Edmonds Dance dataset. All BERT models were trained for 3 epochs, and used a sequence length of 128, batch size of 32, learning rate of 2e \u22125 , and warmup proportion of 0.1. The performance of these models, depicted in Table A4 , were then compared to the performance of baseline Naive Bayes and Random Forest models, shown in Tables A5 and A6. Only the baseline Naive Bayes model trained on augmented CBET data is depicted in Table A5 , as the Naive Bayes model trained on normal CBET data had precision and recall of zero for each emotion.",
"cite_spans": [],
"ref_spans": [
{
"start": 670,
"end": 678,
"text": "Table A4",
"ref_id": "TABREF6"
},
{
"start": 878,
"end": 886,
"text": "Table A5",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "A.5 Evaluation",
"sec_num": null
},
{
"text": "It can be seen from Table A4 that BERT models trained on CBET did not generalize well to lyrical datasets. While models for joy and sadness improved on the performance of Naive Bayes and Random Forest classifiers, models for other emotions did not significantly improve on the baseline, and in some cases performed worse than baseline classifiers. BERT models for anger and fear had lower precision and recall than corresponding to the baseline. Additionally, BERT and Random Forest models were unable to correctly identify disgust, while Naive Bayes models successfully identified multiple instances of disgust. As there was not a significant difference in balance between emotion classes within the CBET dataset, the fact that data augmentation did not significantly improve baseline precision and recall implies that class imbalance was not a main factor in discrepancies between classification accuracy of different emotions.",
"cite_spans": [],
"ref_spans": [
{
"start": 20,
"end": 28,
"text": "Table A4",
"ref_id": "TABREF6"
}
],
"eq_spans": [],
"section": "A.5 Evaluation",
"sec_num": null
},
{
"text": "To compare emotion prediction accuracies across multiple text corpora, we then trained BERT models on the TEC and DailyDialog datasets, and tested them on the Edmonds Dance and Mihalcea/Strapparava datasets. The results are summarized in Tables A7 and A8. Both test accuracy for the TEC and the DailyDialog models were similar to those of the CBET models, implying that the dialog domain does not necessarily show more promise than the social media domain when considering the complex emotion classification problem in lyrics.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "A.5 Evaluation",
"sec_num": null
},
{
"text": "Finally, to compare in-domain model accuracy with our out of domain results, we trained and tested BERT models on the larger, original versions of the Edmonds Dance and Mihalcea/Strapparava datasets respectively, and vice versa. The results are summarized below in Table A9 . We found that the accuracies of models trained and tested on the Edmonds Dance and Mihalcea/Strapparava datasets were on par with those of the out of domain models despite the much smaller training size and genre differences across the lyrical datasets, implying a significant advantage in using in-domain data to train models for complex emotion classification of songs.",
"cite_spans": [],
"ref_spans": [
{
"start": 265,
"end": 273,
"text": "Table A9",
"ref_id": null
}
],
"eq_spans": [],
"section": "A.5 Evaluation",
"sec_num": null
},
{
"text": "It is important to note that precision and recall values for disgust, fear, and surprise remained very low, which could imply that certain emotions are generally more difficult than others to classify. This conclusion is supported by our Turker error analysis in Section 3.1.3, in which we found that emotions such as anticipation, disgust, fear and surprise had relatively lower inter-annotator agreement, while other emotions such as joy and sadness had relatively high agreement.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "A.5 Evaluation",
"sec_num": null
},
{
"text": "Mihalcea and Strapparava included a table in their paper with the number of lines that each of their 6 core emotions was present in, as well as the average magnitude for each emotion across all annotated lines. We used this information to calculate the average magnitude for each emotion across lines in which they were present, shown in Table A10 . As emotions were annotated on a scale from 0",
"cite_spans": [],
"ref_spans": [
{
"start": 338,
"end": 347,
"text": "Table A10",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "A.6 Miscellaneous: Emotion Magnitudes by Line",
"sec_num": null
},
{
"text": "The Edmonds Dance dataset is available by request from the authors of this paper.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "BERT BASE is available at https://tfhub.dev/ google/bert_uncased_L-12_H-768_A-12/1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "to 10, we found it worthwhile to note that annotations for the presence of negative emotions such as anger, disgust, and fear were more likely to be mild than strong. We also found it interesting that only joy had an average magnitude greater than 3, which represented the cutoff for the presence of an emotion .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "annex",
"sec_num": null
}
],
"bib_entries": {
"BIBREF1": {
"ref_id": "b1",
"title": "Emotion classification of song lyrics using bidirectional lstm method with glove word representation weighting",
"authors": [
{
"first": "Jiddy",
"middle": [],
"last": "Abdillah",
"suffix": ""
},
{
"first": "Ibnu",
"middle": [],
"last": "Asror",
"suffix": ""
},
{
"first": "Yanuar Firdaus Arie",
"middle": [],
"last": "Wibowo",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "4",
"issue": "",
"pages": "723--729",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jiddy Abdillah, Ibnu Asror, Yanuar Firdaus Arie Wibowo, et al. 2020. Emotion classification of song lyrics using bidirectional lstm method with glove word representation weighting. Jurnal RESTI (Rekayasa Sistem Dan Teknologi Informasi), 4(4):723-729.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Emotions from text: machine learning for text-based emotion prediction",
"authors": [
{
"first": "Cecilia",
"middle": [],
"last": "Ovesdotter Alm",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Roth",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Sproat",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of human language technology conference and conference on empirical methods in natural language processing",
"volume": "",
"issue": "",
"pages": "579--586",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cecilia Ovesdotter Alm, Dan Roth, and Richard Sproat. 2005. Emotions from text: machine learning for text-based emotion prediction. In Proceedings of human language technology conference and confer- ence on empirical methods in natural language pro- cessing, pages 579-586.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Naive bayes classifiers for music emotion classification based on lyrics",
"authors": [
{
"first": "Yunjing",
"middle": [],
"last": "An",
"suffix": ""
},
{
"first": "Shutao",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "Shujuan",
"middle": [],
"last": "Wang",
"suffix": ""
}
],
"year": 2017,
"venue": "2017 IEEE/ACIS 16th International Conference on Computer and Information Science (ICIS)",
"volume": "",
"issue": "",
"pages": "635--638",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yunjing An, Shutao Sun, and Shujuan Wang. 2017. Naive bayes classifiers for music emotion classifica- tion based on lyrics. In 2017 IEEE/ACIS 16th Inter- national Conference on Computer and Information Science (ICIS), pages 635-638. IEEE.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Interannotator agreement in sentiment analysis: Machine learning perspective",
"authors": [
{
"first": "Victoria",
"middle": [],
"last": "Bobicev",
"suffix": ""
},
{
"first": "Marina",
"middle": [],
"last": "Sokolova",
"suffix": ""
}
],
"year": 2017,
"venue": "RANLP",
"volume": "",
"issue": "",
"pages": "97--102",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Victoria Bobicev and Marina Sokolova. 2017. Inter- annotator agreement in sentiment analysis: Machine learning perspective. In RANLP, pages 97-102.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Emobank: Studying the impact of annotation perspective and representation format on dimensional emotion analysis",
"authors": [
{
"first": "Sven",
"middle": [],
"last": "Buechel",
"suffix": ""
},
{
"first": "Udo",
"middle": [],
"last": "Hahn",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 15th Conference of the European Chapter",
"volume": "2",
"issue": "",
"pages": "578--585",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sven Buechel and Udo Hahn. 2017. Emobank: Study- ing the impact of annotation perspective and repre- sentation format on dimensional emotion analysis. In Proceedings of the 15th Conference of the Euro- pean Chapter of the Association for Computational Linguistics: Volume 2, Short Papers, pages 578- 585.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Moodylyrics: A sentiment annotated lyrics dataset",
"authors": [
{
"first": "Erion",
"middle": [],
"last": "\u00c7ano",
"suffix": ""
},
{
"first": "Maurizio",
"middle": [],
"last": "Morisio",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 2017 International Conference on Intelligent Systems, Metaheuristics & Swarm Intelligence",
"volume": "",
"issue": "",
"pages": "118--124",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Erion \u00c7ano and Maurizio Morisio. 2017. Moodylyrics: A sentiment annotated lyrics dataset. In Proceed- ings of the 2017 International Conference on Intelli- gent Systems, Metaheuristics & Swarm Intelligence, pages 118-124.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Using random forest to learn imbalanced data",
"authors": [
{
"first": "Chao",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Andy",
"middle": [],
"last": "Liaw",
"suffix": ""
},
{
"first": "Leo",
"middle": [],
"last": "Breiman",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "110",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chao Chen, Andy Liaw, Leo Breiman, et al. 2004. Us- ing random forest to learn imbalanced data. Univer- sity of California, Berkeley, 110(1-12):24.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Music mood detection based on audio and lyrics with deep neural net",
"authors": [
{
"first": "R\u00e9mi",
"middle": [],
"last": "Delbouys",
"suffix": ""
},
{
"first": "Romain",
"middle": [],
"last": "Hennequin",
"suffix": ""
},
{
"first": "Francesco",
"middle": [],
"last": "Piccoli",
"suffix": ""
},
{
"first": "Jimena",
"middle": [],
"last": "Royo-Letelier",
"suffix": ""
},
{
"first": "Manuel",
"middle": [],
"last": "Moussallam",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 19th International Society for Music Information Retrieval Conference",
"volume": "",
"issue": "",
"pages": "370--375",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R\u00e9mi Delbouys, Romain Hennequin, Francesco Pic- coli, Jimena Royo-Letelier, and Manuel Moussal- lam. 2018. Music mood detection based on audio and lyrics with deep neural net. In Proceedings of the 19th International Society for Music Information Retrieval Conference, ISMIR 2018, Paris, France, September 23-27, 2018, pages 370-375.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "BERT: Pre-training of deep bidirectional transformers for language understanding",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "1",
"issue": "",
"pages": "4171--4186",
"other_ids": {
"DOI": [
"10.18653/v1/N19-1423"
]
},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language under- standing. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171-4186, Minneapolis, Minnesota. Associ- ation for Computational Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Characterizing debate performance via aggregated twitter sentiment",
"authors": [
{
"first": "A",
"middle": [],
"last": "Nicholas",
"suffix": ""
},
{
"first": "David",
"middle": [
"A"
],
"last": "Diakopoulos",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Shamma",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the SIGCHI conference on human factors in computing systems",
"volume": "",
"issue": "",
"pages": "1195--1198",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nicholas A Diakopoulos and David A Shamma. 2010. Characterizing debate performance via aggregated twitter sentiment. In Proceedings of the SIGCHI conference on human factors in computing systems, pages 1195-1198.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "The threatening nature of \"rap\" music. Psychology",
"authors": [
{
"first": "Adam",
"middle": [],
"last": "Dunbar",
"suffix": ""
},
{
"first": "Charis",
"middle": [
"E"
],
"last": "Kubrin",
"suffix": ""
},
{
"first": "Nicholas",
"middle": [],
"last": "Scurich",
"suffix": ""
}
],
"year": 2016,
"venue": "Public Policy, and Law",
"volume": "22",
"issue": "3",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Adam Dunbar, Charis E Kubrin, and Nicholas Scurich. 2016. The threatening nature of \"rap\" music. Psy- chology, Public Policy, and Law, 22(3):280.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Facial expression and emotion",
"authors": [
{
"first": "Paul",
"middle": [],
"last": "Ekman",
"suffix": ""
}
],
"year": 1993,
"venue": "American psychologist",
"volume": "48",
"issue": "4",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Paul Ekman. 1993. Facial expression and emotion. American psychologist, 48(4):384.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Unmasking the face: A guide to recognizing emotions from facial clues",
"authors": [
{
"first": "Paul",
"middle": [],
"last": "Ekman",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Wallace V Friesen",
"suffix": ""
}
],
"year": 2003,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Paul Ekman and Wallace V Friesen. 2003. Unmask- ing the face: A guide to recognizing emotions from facial clues. ISHK.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Who's afraid of rap: Differential reactions to music lyrics 1",
"authors": [
{
"first": "B",
"middle": [],
"last": "Carrie",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Fried",
"suffix": ""
}
],
"year": 1999,
"venue": "Journal of Applied Social Psychology",
"volume": "29",
"issue": "4",
"pages": "705--721",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Carrie B Fried. 1999. Who's afraid of rap: Differen- tial reactions to music lyrics 1. Journal of Applied Social Psychology, 29(4):705-721.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Using hashtags as labels for supervised learning of emotions in twitter messages",
"authors": [
{
"first": "Maryam",
"middle": [],
"last": "Hasan",
"suffix": ""
},
{
"first": "Emmanuel",
"middle": [],
"last": "Agu",
"suffix": ""
},
{
"first": "Elke",
"middle": [],
"last": "Rundensteiner",
"suffix": ""
}
],
"year": 2014,
"venue": "ACM SIGKDD workshop on health informatics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Maryam Hasan, Emmanuel Agu, and Elke Runden- steiner. 2014. Using hashtags as labels for super- vised learning of emotions in twitter messages. In ACM SIGKDD workshop on health informatics, New York, USA.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Sentiment analysis of political communication: combining a dictionary approach with crowdcoding. Quality & quantity",
"authors": [
{
"first": "Martin",
"middle": [],
"last": "Haselmayer",
"suffix": ""
},
{
"first": "Marcelo",
"middle": [],
"last": "Jenny",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "51",
"issue": "",
"pages": "2623--2646",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Martin Haselmayer and Marcelo Jenny. 2017. Senti- ment analysis of political communication: combin- ing a dictionary approach with crowdcoding. Qual- ity & quantity, 51(6):2623-2646.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Language feature mining for music emotion classification via supervised learning from lyrics",
"authors": [
{
"first": "Hui",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Jianming",
"middle": [],
"last": "Jin",
"suffix": ""
},
{
"first": "Yuhong",
"middle": [],
"last": "Xiong",
"suffix": ""
},
{
"first": "Bo",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Wu",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "Ling",
"middle": [],
"last": "Zhao",
"suffix": ""
}
],
"year": 2008,
"venue": "International Symposium on Intelligence Computation and Applications",
"volume": "",
"issue": "",
"pages": "426--435",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hui He, Jianming Jin, Yuhong Xiong, Bo Chen, Wu Sun, and Ling Zhao. 2008. Language feature mining for music emotion classification via super- vised learning from lyrics. In International Sympo- sium on Intelligence Computation and Applications, pages 426-435. Springer.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "When lyrics outperform audio for music mood classification: A feature analysis",
"authors": [
{
"first": "Xiao",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Stephen Downie",
"suffix": ""
}
],
"year": 2010,
"venue": "ISMIR",
"volume": "",
"issue": "",
"pages": "619--624",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiao Hu and J Stephen Downie. 2010. When lyrics outperform audio for music mood classification: A feature analysis. In ISMIR, pages 619-624.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Lyric text mining in music mood classification",
"authors": [
{
"first": "Xiao",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "Andreas",
"middle": [
"F"
],
"last": "Stephen Downie",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Ehmann",
"suffix": ""
}
],
"year": 2009,
"venue": "",
"volume": "183",
"issue": "",
"pages": "2--209",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiao Hu, J Stephen Downie, and Andreas F Ehmann. 2009. Lyric text mining in music mood classifica- tion. American music, 183(5,049):2-209.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Emotion analysis of songs based on lyrical and audio features",
"authors": [
{
"first": "Adit",
"middle": [],
"last": "Jamdar",
"suffix": ""
},
{
"first": "Jessica",
"middle": [],
"last": "Abraham",
"suffix": ""
},
{
"first": "Karishma",
"middle": [],
"last": "Khanna",
"suffix": ""
},
{
"first": "Rahul",
"middle": [],
"last": "Dubey",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1506.05012"
]
},
"num": null,
"urls": [],
"raw_text": "Adit Jamdar, Jessica Abraham, Karishma Khanna, and Rahul Dubey. 2015. Emotion analysis of songs based on lyrical and audio features. arXiv preprint arXiv:1506.05012.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "On negative results when using sentiment analysis tools for software engineering research",
"authors": [
{
"first": "Robbert",
"middle": [],
"last": "Jongeling",
"suffix": ""
},
{
"first": "Proshanta",
"middle": [],
"last": "Sarkar",
"suffix": ""
},
{
"first": "Subhajit",
"middle": [],
"last": "Datta",
"suffix": ""
},
{
"first": "Alexander",
"middle": [],
"last": "Serebrenik",
"suffix": ""
}
],
"year": 2017,
"venue": "Empirical Software Engineering",
"volume": "22",
"issue": "5",
"pages": "2543--2584",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robbert Jongeling, Proshanta Sarkar, Subhajit Datta, and Alexander Serebrenik. 2017. On negative re- sults when using sentiment analysis tools for soft- ware engineering research. Empirical Software En- gineering, 22(5):2543-2584.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Lyrics-based emotion classification using feature selection by partial syntactic analysis",
"authors": [
{
"first": "Minho",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "Hyuk-Chul",
"middle": [],
"last": "Kwon",
"suffix": ""
}
],
"year": 2011,
"venue": "2011 IEEE 23rd International Conference on Tools with Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "960--964",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Minho Kim and Hyuk-Chul Kwon. 2011. Lyrics-based emotion classification using feature selection by par- tial syntactic analysis. In 2011 IEEE 23rd Inter- national Conference on Tools with Artificial Intelli- gence, pages 960-964. IEEE.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Music emotion recognition: A state of the art review",
"authors": [
{
"first": "E",
"middle": [],
"last": "Youngmoo",
"suffix": ""
},
{
"first": "Erik",
"middle": [
"M"
],
"last": "Kim",
"suffix": ""
},
{
"first": "Raymond",
"middle": [],
"last": "Schmidt",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Migneco",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Brandon",
"suffix": ""
},
{
"first": "Patrick",
"middle": [],
"last": "Morton",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Richardson",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Scott",
"suffix": ""
}
],
"year": 2010,
"venue": "Proc. ismir",
"volume": "86",
"issue": "",
"pages": "937--952",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Youngmoo E Kim, Erik M Schmidt, Raymond Migneco, Brandon G Morton, Patrick Richardson, Jeffrey Scott, Jacquelin A Speck, and Douglas Turn- bull. 2010. Music emotion recognition: A state of the art review. In Proc. ismir, volume 86, pages 937- 952.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Dailydialog: A manually labelled multi-turn dialogue dataset",
"authors": [
{
"first": "Yanran",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Hui",
"middle": [],
"last": "Su",
"suffix": ""
},
{
"first": "Xiaoyu",
"middle": [],
"last": "Shen",
"suffix": ""
},
{
"first": "Wenjie",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Ziqiang",
"middle": [],
"last": "Cao",
"suffix": ""
},
{
"first": "Shuzi",
"middle": [],
"last": "Niu",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the Eighth International Joint Conference on Natural Language Processing",
"volume": "1",
"issue": "",
"pages": "986--995",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yanran Li, Hui Su, Xiaoyu Shen, Wenjie Li, Ziqiang Cao, and Shuzi Niu. 2017. Dailydialog: A manually labelled multi-turn dialogue dataset. In Proceedings of the Eighth International Joint Conference on Nat- ural Language Processing (Volume 1: Long Papers), pages 986-995.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Research on multimodal music emotion classification based on audio and lyirc",
"authors": [
{
"first": "Gaojun",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Zhiyuan",
"middle": [],
"last": "Tan",
"suffix": ""
}
],
"year": 2020,
"venue": "2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC)",
"volume": "1",
"issue": "",
"pages": "2331--2335",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gaojun Liu and Zhiyuan Tan. 2020. Research on multi- modal music emotion classification based on audio and lyirc. In 2020 IEEE 4th Information Technol- ogy, Networking, Electronic and Automation Con- trol Conference (ITNEC), volume 1, pages 2331- 2335. IEEE.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Lyrics, music, and emotions",
"authors": [
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "Carlo",
"middle": [],
"last": "Strapparava",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 2012",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rada Mihalcea and Carlo Strapparava. 2012. Lyrics, music, and emotions. In Proceedings of the 2012",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "Natural Language Processing and Computational Natural Language Learning",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "590--599",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 590-599.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "# emotional tweets",
"authors": [
{
"first": "Saif",
"middle": [],
"last": "Mohammad",
"suffix": ""
}
],
"year": 2012,
"venue": "* SEM 2012: The First Joint Conference on Lexical and Computational Semantics",
"volume": "1",
"issue": "",
"pages": "246--255",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Saif Mohammad. 2012. # emotional tweets. In * SEM 2012: The First Joint Conference on Lexical and Computational Semantics-Volume 1: Proceedings of the main conference and the shared task, and Vol- ume 2: Proceedings of the Sixth International Work- shop on Semantic Evaluation (SemEval 2012), pages 246-255.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "Emotion intensities in tweets",
"authors": [
{
"first": "Saif",
"middle": [],
"last": "Mohammad",
"suffix": ""
},
{
"first": "Felipe",
"middle": [],
"last": "Bravo-Marquez",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 6th Joint Conference on Lexical and Computational Semantics (*SEM 2017)",
"volume": "",
"issue": "",
"pages": "65--77",
"other_ids": {
"DOI": [
"10.18653/v1/S17-1007"
]
},
"num": null,
"urls": [],
"raw_text": "Saif Mohammad and Felipe Bravo-Marquez. 2017. Emotion intensities in tweets. In Proceedings of the 6th Joint Conference on Lexical and Computational Semantics (*SEM 2017), pages 65-77, Vancouver, Canada. Association for Computational Linguistics.",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "Crowdsourcing a word-emotion association lexicon",
"authors": [
{
"first": "M",
"middle": [],
"last": "Saif",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Mohammad",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Peter",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Turney",
"suffix": ""
}
],
"year": 2013,
"venue": "Computational Intelligence",
"volume": "29",
"issue": "3",
"pages": "436--465",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Saif M Mohammad and Peter D Turney. 2013. Crowd- sourcing a word-emotion association lexicon. Com- putational Intelligence, 29(3):436-465.",
"links": null
},
"BIBREF31": {
"ref_id": "b31",
"title": "An analysis of annotated corpora for emotion classification in text",
"authors": [
{
"first": "Laura",
"middle": [
"Ana"
],
"last": "",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Oberl\u00e4nder",
"suffix": ""
},
{
"first": "Roman",
"middle": [],
"last": "Klinger",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 27th International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "2104--2119",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Laura Ana Maria Oberl\u00e4nder and Roman Klinger. 2018. An analysis of annotated corpora for emotion clas- sification in text. In Proceedings of the 27th Inter- national Conference on Computational Linguistics, pages 2104-2119.",
"links": null
},
"BIBREF32": {
"ref_id": "b32",
"title": "Exploiting synchronized lyrics and vocal features for music emotion detection",
"authors": [
{
"first": "Loreto",
"middle": [],
"last": "Parisi",
"suffix": ""
},
{
"first": "Simone",
"middle": [],
"last": "Francia",
"suffix": ""
},
{
"first": "Silvio",
"middle": [],
"last": "Olivastri",
"suffix": ""
},
{
"first": "Maria",
"middle": [
"Stella"
],
"last": "Tavella",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1901.04831"
]
},
"num": null,
"urls": [],
"raw_text": "Loreto Parisi, Simone Francia, Silvio Olivastri, and Maria Stella Tavella. 2019. Exploiting synchronized lyrics and vocal features for music emotion detec- tion. arXiv preprint arXiv:1901.04831.",
"links": null
},
"BIBREF33": {
"ref_id": "b33",
"title": "The nature of emotions: Human emotions have deep evolutionary roots, a fact that may explain their complexity and provide tools for clinical practice",
"authors": [
{
"first": "Robert",
"middle": [],
"last": "Plutchik",
"suffix": ""
}
],
"year": 2001,
"venue": "American scientist",
"volume": "89",
"issue": "4",
"pages": "344--350",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robert Plutchik. 2001. The nature of emotions: Hu- man emotions have deep evolutionary roots, a fact that may explain their complexity and provide tools for clinical practice. American scientist, 89(4):344- 350.",
"links": null
},
"BIBREF34": {
"ref_id": "b34",
"title": "Refining the most frequent sense baseline",
"authors": [
{
"first": "Judita",
"middle": [],
"last": "Preiss",
"suffix": ""
},
{
"first": "Jon",
"middle": [],
"last": "Dehdari",
"suffix": ""
},
{
"first": "Josh",
"middle": [],
"last": "King",
"suffix": ""
},
{
"first": "Dennis",
"middle": [],
"last": "Mehay",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Workshop on Semantic Evaluations: Recent Achievements and Future Directions",
"volume": "",
"issue": "",
"pages": "10--18",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Judita Preiss, Jon Dehdari, Josh King, and Dennis Mehay. 2009. Refining the most frequent sense base- line. In Proceedings of the Workshop on Semantic Evaluations: Recent Achievements and Future Di- rections (SEW-2009), pages 10-18.",
"links": null
},
"BIBREF35": {
"ref_id": "b35",
"title": "Validating automated sentiment analysis of online cognitive behavioral therapy patient texts: An exploratory study",
"authors": [
{
"first": "Simon",
"middle": [],
"last": "Provoost",
"suffix": ""
},
{
"first": "Jeroen",
"middle": [],
"last": "Ruwaard",
"suffix": ""
},
{
"first": "Heleen",
"middle": [],
"last": "Ward Van Breda",
"suffix": ""
},
{
"first": "Tibor",
"middle": [],
"last": "Riper",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Bosse",
"suffix": ""
}
],
"year": 2019,
"venue": "Frontiers in psychology",
"volume": "10",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Simon Provoost, Jeroen Ruwaard, Ward van Breda, Heleen Riper, and Tibor Bosse. 2019. Validating automated sentiment analysis of online cognitive be- havioral therapy patient texts: An exploratory study. Frontiers in psychology, 10:1065.",
"links": null
},
"BIBREF36": {
"ref_id": "b36",
"title": "Music emotion classification based on lyrics-audio using corpus based emotion",
"authors": [
{
"first": "Riyanarto",
"middle": [],
"last": "Fika Hastarita Rachman",
"suffix": ""
},
{
"first": "Chastine",
"middle": [],
"last": "Sarno",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Fatichah",
"suffix": ""
}
],
"year": 2018,
"venue": "International Journal of Electrical & Computer Engineering",
"volume": "8",
"issue": "3",
"pages": "2088--8708",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Fika Hastarita Rachman, Riyanarto Sarno, and Chas- tine Fatichah. 2018. Music emotion classification based on lyrics-audio using corpus based emotion. International Journal of Electrical & Computer En- gineering (2088-8708), 8(3).",
"links": null
},
"BIBREF37": {
"ref_id": "b37",
"title": "Online sentiment analysis in marketing research: a review",
"authors": [
{
"first": "Meena",
"middle": [],
"last": "Rambocas",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Barney",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Pacheco",
"suffix": ""
}
],
"year": 2018,
"venue": "Journal of Research in Interactive Marketing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Meena Rambocas and Barney G Pacheco. 2018. On- line sentiment analysis in marketing research: a re- view. Journal of Research in Interactive Marketing.",
"links": null
},
"BIBREF38": {
"ref_id": "b38",
"title": "Naive bayes and text classification i-introduction and theory",
"authors": [
{
"first": "Sebastian",
"middle": [],
"last": "Raschka",
"suffix": ""
}
],
"year": 2014,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1410.5329"
]
},
"num": null,
"urls": [],
"raw_text": "Sebastian Raschka. 2014. Naive bayes and text clas- sification i-introduction and theory. arXiv preprint arXiv:1410.5329.",
"links": null
},
"BIBREF39": {
"ref_id": "b39",
"title": "Annotation, modelling and analysis of fine-grained emotions on a stance and sentiment detection corpus",
"authors": [
{
"first": "Hendrik",
"middle": [],
"last": "Schuff",
"suffix": ""
},
{
"first": "Jeremy",
"middle": [],
"last": "Barnes",
"suffix": ""
},
{
"first": "Julian",
"middle": [],
"last": "Mohme",
"suffix": ""
},
{
"first": "Sebastian",
"middle": [],
"last": "Pad\u00f3",
"suffix": ""
},
{
"first": "Roman",
"middle": [],
"last": "Klinger",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis",
"volume": "",
"issue": "",
"pages": "13--23",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hendrik Schuff, Jeremy Barnes, Julian Mohme, Sebas- tian Pad\u00f3, and Roman Klinger. 2017. Annotation, modelling and analysis of fine-grained emotions on a stance and sentiment detection corpus. In Pro- ceedings of the 8th Workshop on Computational Ap- proaches to Subjectivity, Sentiment and Social Me- dia Analysis, pages 13-23.",
"links": null
},
"BIBREF40": {
"ref_id": "b40",
"title": "Emotion detection from text",
"authors": [
{
"first": "Ameneh Gholipour",
"middle": [],
"last": "Shahraki",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ameneh Gholipour Shahraki. 2015. Emotion detection from text. Master's thesis, University of Alberta.",
"links": null
},
"BIBREF41": {
"ref_id": "b41",
"title": "A parallel corpus of music and lyrics annotated with emotions",
"authors": [
{
"first": "Carlo",
"middle": [],
"last": "Strapparava",
"suffix": ""
},
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "Alberto",
"middle": [],
"last": "Battocchi",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)",
"volume": "",
"issue": "",
"pages": "2343--2346",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Carlo Strapparava, Rada Mihalcea, and Alberto Battoc- chi. 2012. A parallel corpus of music and lyrics an- notated with emotions. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12), pages 2343-2346, Istan- bul, Turkey. European Language Resources Associ- ation (ELRA).",
"links": null
},
"BIBREF42": {
"ref_id": "b42",
"title": "Cultural stereotyping of emotional responses to music genre",
"authors": [
{
"first": "Marco",
"middle": [],
"last": "Susino",
"suffix": ""
},
{
"first": "Emery",
"middle": [],
"last": "Schubert",
"suffix": ""
}
],
"year": 2019,
"venue": "Psychology of Music",
"volume": "47",
"issue": "3",
"pages": "342--357",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marco Susino and Emery Schubert. 2019a. Cultural stereotyping of emotional responses to music genre. Psychology of Music, 47(3):342-357.",
"links": null
},
"BIBREF43": {
"ref_id": "b43",
"title": "Negative emotion responses to heavy-metal and hip-hop music with positive lyrics",
"authors": [
{
"first": "Marco",
"middle": [],
"last": "Susino",
"suffix": ""
},
{
"first": "Emery",
"middle": [],
"last": "Schubert",
"suffix": ""
}
],
"year": 2019,
"venue": "Empirical Musicology Review",
"volume": "14",
"issue": "1-2",
"pages": "2--15",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marco Susino and Emery Schubert. 2019b. Negative emotion responses to heavy-metal and hip-hop mu- sic with positive lyrics. Empirical Musicology Re- view, 14(1-2):2-15.",
"links": null
},
"BIBREF44": {
"ref_id": "b44",
"title": "Gold-standard for topic-specific sentiment analysis of economic texts",
"authors": [
{
"first": "Pyry",
"middle": [],
"last": "Takala",
"suffix": ""
},
{
"first": "Pekka",
"middle": [],
"last": "Malo",
"suffix": ""
},
{
"first": "Ankur",
"middle": [],
"last": "Sinha",
"suffix": ""
},
{
"first": "Oskar",
"middle": [],
"last": "Ahlgren",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)",
"volume": "",
"issue": "",
"pages": "2152--2157",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pyry Takala, Pekka Malo, Ankur Sinha, and Oskar Ahlgren. 2014. Gold-standard for topic-specific sen- timent analysis of economic texts. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14), pages 2152- 2157, Reykjavik, Iceland. European Language Re- sources Association (ELRA).",
"links": null
},
"BIBREF45": {
"ref_id": "b45",
"title": "Music emotion classification of chinese songs based on lyrics using tf* idf and rhyme",
"authors": [
{
"first": "Xing",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Xiaoou",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Deshun",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Yuqian",
"middle": [],
"last": "Wu",
"suffix": ""
}
],
"year": 2011,
"venue": "ISMIR",
"volume": "",
"issue": "",
"pages": "765--770",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xing Wang, Xiaoou Chen, Deshun Yang, and Yuqian Wu. 2011. Music emotion classification of chinese songs based on lyrics using tf* idf and rhyme. In ISMIR, pages 765-770. Citeseer.",
"links": null
},
"BIBREF46": {
"ref_id": "b46",
"title": "Music emotion identification from lyrics",
"authors": [
{
"first": "Dan",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Won-Sook",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2009,
"venue": "11th IEEE International Symposium on Multimedia",
"volume": "",
"issue": "",
"pages": "624--629",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dan Yang and Won-Sook Lee. 2009. Music emotion identification from lyrics. In 2009 11th IEEE Inter- national Symposium on Multimedia, pages 624-629. IEEE.",
"links": null
},
"BIBREF47": {
"ref_id": "b47",
"title": "Toward multi-modal music emotion classification",
"authors": [
{
"first": "Yi-Hsuan",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Yu-Ching",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Heng-Tze",
"middle": [],
"last": "Cheng",
"suffix": ""
},
{
"first": "I-Bin",
"middle": [],
"last": "Liao",
"suffix": ""
},
{
"first": "Yeh-Chin",
"middle": [],
"last": "Ho",
"suffix": ""
},
{
"first": "Homer H",
"middle": [],
"last": "Chen",
"suffix": ""
}
],
"year": 2008,
"venue": "Pacific-Rim Conference on Multimedia",
"volume": "",
"issue": "",
"pages": "70--79",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yi-Hsuan Yang, Yu-Ching Lin, Heng-Tze Cheng, I-Bin Liao, Yeh-Chin Ho, and Homer H Chen. 2008. To- ward multi-modal music emotion classification. In Pacific-Rim Conference on Multimedia, pages 70- 79. Springer.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"uris": null,
"text": "Genres within Mihalcea/Strapparava HITs",
"type_str": "figure",
"num": null
},
"FIGREF1": {
"uris": null,
"text": "Figure A1: HIT Preliminary Instructions",
"type_str": "figure",
"num": null
},
"FIGREF3": {
"uris": null,
"text": "Figure A6: Annotator Enjoyment/Familiarity of Songs by Count and Annotation Quality",
"type_str": "figure",
"num": null
},
"TABREF1": {
"num": null,
"content": "<table><tr><td/><td/><td>: Examples from the Edmonds Dance Dataset</td></tr><tr><td/><td colspan=\"2\">Mihalcea/Strapparava Edmonds Dance</td></tr><tr><td>Songs</td><td>100</td><td>524</td></tr><tr><td>Lines</td><td>4976</td><td>22924</td></tr><tr><td>Words</td><td>109332</td><td>708985</td></tr><tr><td>Vocabulary</td><td>2233</td><td>6563</td></tr></table>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF2": {
"num": null,
"content": "<table><tr><td>: Lyrical Dataset Basic Statistics</td></tr><tr><td>erating 452 verse-based and 100 song-based anno-</td></tr><tr><td>tations. Table 2 provides some basic statistics for</td></tr><tr><td>the lyrical datasets used in our research.</td></tr><tr><td>Mechanical Turk We submitted HITs on Me-</td></tr><tr><td>chanical Turk to validate lyric annotations. Each</td></tr><tr><td>HIT contained three songs to be annotated from</td></tr><tr><td>the reader's perspective for 8 emotions on a 6-point</td></tr><tr><td>Likert scale. We also queried whether the annotator</td></tr><tr><td>had heard of each song (yes/no), and whether they</td></tr><tr><td>liked it (yes/no/unsure). Of the 186 songs anno-</td></tr><tr><td>tated in total, 93 were from the Edmonds Dance</td></tr><tr><td>dataset and 93 were from the Mihalcea/Strapparava</td></tr><tr><td>dataset. HITs encompassed multiple genres, with</td></tr><tr><td>the Edmonds Dance dataset mostly consisting of</td></tr><tr><td>electronic music and the Mihalcea/Strapparava</td></tr><tr><td>dataset mostly consisting of rock music. Figures 1</td></tr><tr><td>and 2 summarize HIT breakdowns by genre.</td></tr></table>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF4": {
"num": null,
"content": "<table/>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF5": {
"num": null,
"content": "<table><tr><td colspan=\"2\">Emotion CBET TEC</td><td>DD</td><td>Dance</td><td>M/S</td></tr><tr><td>Anger</td><td>11.2% 7.4%</td><td colspan=\"3\">1.0% 13.7% 9.1%</td></tr><tr><td colspan=\"2\">Disgust 10.7% 3.6%</td><td colspan=\"3\">0.3% 21.9% 2.9%</td></tr><tr><td>Fear</td><td colspan=\"4\">11.2% 13.3% 0.2% 19.7% 1.8%</td></tr><tr><td>Joy</td><td colspan=\"4\">13.4% 39.1% 12.5% 43.9% 50.4%</td></tr><tr><td colspan=\"5\">Sadness 11.4% 18.2% 1.1% 35.3% 33.0%</td></tr><tr><td colspan=\"5\">Surprise 11.4% 18.3% 1.8% 13.0% 0.9%</td></tr></table>",
"html": null,
"text": "datasets, three large collections of text annotated for multiple emotions including 6 core emotions present in both the Edmonds Dance and Mihalcea/Strapparava datasets. The CBET and TEC datasets respectively consist of 81,163 and 21,048 tweets, while the DailyDialog dataset consists of 102,979 statements collected from 13,118 transcripts of two-person conversations. Emotion distributions of the CBET, TEC, Daily Dialog, Edmonds Dance, and Mihalcea/Strapparava datasets are depicted inTable 4.",
"type_str": "table"
},
"TABREF6": {
"num": null,
"content": "<table/>",
"html": null,
"text": "Presence of Emotion by Dataset",
"type_str": "table"
},
"TABREF8": {
"num": null,
"content": "<table><tr><td colspan=\"4\">Emotion Naive Bayes CBET BERT Lyrics BERT</td></tr><tr><td>Anger</td><td>0.17</td><td>0.04</td><td>0.2</td></tr><tr><td>Disgust</td><td>0.21</td><td>0</td><td>0</td></tr><tr><td>Fear</td><td>0.18</td><td>0.14</td><td>0</td></tr><tr><td>Joy</td><td>0.03</td><td>0.24</td><td>0.69</td></tr><tr><td>Sadness</td><td>0.55</td><td>0.48</td><td>0.54</td></tr><tr><td>Surprise</td><td>0</td><td>0</td><td>0</td></tr></table>",
"html": null,
"text": "Model Accuracy on Lyrics By Emotion",
"type_str": "table"
},
"TABREF9": {
"num": null,
"content": "<table/>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF11": {
"num": null,
"content": "<table><tr><td>Dataset</td><td>Emotion</td><td>Pearson's Correlation</td><td>90% CI</td></tr><tr><td>Dance</td><td>All</td><td>0.396</td><td>(0.343, 0.445)</td></tr><tr><td>Dance</td><td>Anger</td><td>0.204</td><td>(0.033, 0.363)</td></tr><tr><td colspan=\"2\">Dance Anticipation</td><td>0.294</td><td>(0.129, 0.443)</td></tr><tr><td>Dance</td><td>Disgust</td><td>0.429</td><td>(0.278, 0.559)</td></tr><tr><td>Dance</td><td>Fear</td><td>0.31</td><td>(0.146, 0.457)</td></tr><tr><td>Dance</td><td>Joy</td><td>0.362</td><td>(0.203, 0.502)</td></tr><tr><td>Dance</td><td>Sadness</td><td>0.316</td><td>(0.154, 0.462)</td></tr><tr><td>Dance</td><td>Surprise</td><td>0.175</td><td>(0.003, 0.336)</td></tr><tr><td>Dance</td><td>Trust</td><td>0.384</td><td>(0.228, 0.522)</td></tr><tr><td>M/S</td><td>All</td><td>0.183</td><td>(0.124, 0.241)</td></tr><tr><td>M/S</td><td>Anger</td><td>0.28</td><td>(0.114, 0.431)</td></tr><tr><td>M/S</td><td>Disgust</td><td>0.214</td><td>(0.045, 0.371)</td></tr><tr><td>M/S</td><td>Fear</td><td>0.499</td><td>(0.358, 0.618)</td></tr><tr><td>M/S</td><td>Joy</td><td>0.439</td><td>(0.289, 0.568)</td></tr><tr><td>M/S</td><td>Sadness</td><td>0.477</td><td>(0.333, 0.6)</td></tr><tr><td>M/S</td><td>Surprise</td><td>0.01</td><td>(-0.161, 0.18)</td></tr><tr><td>: Interannotator Agreement at Cohen's Kappa</td><td/><td/><td/></tr><tr><td>Threshold of 0.25</td><td/><td/><td/></tr></table>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF12": {
"num": null,
"content": "<table><tr><td>: Pearson's Correlation between Original and</td></tr><tr><td>Turker annotations</td></tr></table>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF14": {
"num": null,
"content": "<table><tr><td>: Krippendorf's Alpha Values By Emotion</td></tr><tr><td>Over 50 Annotator Groups</td></tr><tr><td>Figure A4: Correlation Heatmap of Krippendorf's Al-</td></tr><tr><td>phas</td></tr></table>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF16": {
"num": null,
"content": "<table><tr><td>Emotion</td><td>Test</td><td colspan=\"3\">Accuracy Precision Recall</td></tr><tr><td>Anger</td><td>Dance Original</td><td>0.65</td><td>0.14</td><td>0.29</td></tr><tr><td>Anger</td><td>Dance Turk</td><td>0.63</td><td>0.14</td><td>0.31</td></tr><tr><td>Anger</td><td>M/S Original</td><td>0.7</td><td>0.12</td><td>0.34</td></tr><tr><td>Anger</td><td>M/S Turk</td><td>0.84</td><td>0.23</td><td>0.38</td></tr><tr><td colspan=\"2\">Disgust Dance Original</td><td>0.77</td><td>0.43</td><td>0.2</td></tr><tr><td>Disgust</td><td>Dance Turk</td><td>0.84</td><td>0</td><td>0</td></tr><tr><td>Disgust</td><td>M/S Original</td><td>0.82</td><td>0.04</td><td>0.23</td></tr><tr><td>Disgust</td><td>M/S Turk</td><td>0.92</td><td>0.29</td><td>0.5</td></tr><tr><td>Fear</td><td>Dance Original</td><td>0.7</td><td>0.27</td><td>0.3</td></tr><tr><td>Fear</td><td>Dance Turk</td><td>0.73</td><td>0.35</td><td>0.53</td></tr><tr><td>Fear</td><td>M/S Original</td><td>0.84</td><td>0.03</td><td>0.25</td></tr><tr><td>Fear</td><td>M/S Turk</td><td>0.77</td><td>0.14</td><td>0.18</td></tr><tr><td>Joy</td><td>Dance Original</td><td>0.57</td><td>0.8</td><td>0.02</td></tr><tr><td>Joy</td><td>Dance Turk</td><td>0.45</td><td>0</td><td>0</td></tr><tr><td>Joy</td><td>M/S Original</td><td>0.5</td><td>0.67</td><td>0.02</td></tr><tr><td>Joy</td><td>M/S Turk</td><td>0.54</td><td>0</td><td>0</td></tr><tr><td colspan=\"2\">Sadness Dance Original</td><td>0.61</td><td>0.47</td><td>0.77</td></tr><tr><td>Sadness</td><td>Dance Turk</td><td>0.53</td><td>0.42</td><td>0.79</td></tr><tr><td>Sadness</td><td>M/S Original</td><td>0.55</td><td>0.4</td><td>0.77</td></tr><tr><td>Sadness</td><td>M/S Turk</td><td>0.61</td><td>0.58</td><td>0.71</td></tr><tr><td colspan=\"2\">Surprise Dance Original</td><td>0.86</td><td>0</td><td>0</td></tr><tr><td>Surprise</td><td>Dance Turk</td><td>0.87</td><td>0</td><td>0</td></tr><tr><td>Surprise</td><td>M/S Original</td><td>0.97</td><td>0</td><td>0</td></tr><tr><td>Surprise</td><td>M/S Turk</td><td>0.91</td><td>0</td><td>0</td></tr></table>",
"html": null,
"text": "BERT Trained on CBET Variations",
"type_str": "table"
},
"TABREF17": {
"num": null,
"content": "<table><tr><td>Emotion</td><td>Test</td><td colspan=\"3\">Accuracy Precision Recall</td></tr><tr><td>Anger</td><td>Dance Original</td><td>0.85</td><td>0.1</td><td>0.01</td></tr><tr><td>Anger</td><td>Dance Turk</td><td>0.84</td><td>0</td><td>0</td></tr><tr><td>Anger</td><td>M/S Original</td><td>0.86</td><td>0.04</td><td>0.02</td></tr><tr><td>Anger</td><td>M/S Turk</td><td>0.87</td><td>0</td><td>0</td></tr><tr><td colspan=\"2\">Disgust Dance Original</td><td>0.78</td><td>0</td><td>0</td></tr><tr><td>Disgust</td><td>Dance Turk</td><td>0.95</td><td>0</td><td>0</td></tr><tr><td>Disgust</td><td>M/S Original</td><td>0.96</td><td>0</td><td>0</td></tr><tr><td>Disgust</td><td>M/S Turk</td><td>0.96</td><td>0</td><td>0</td></tr><tr><td>Fear</td><td>Dance Original</td><td>0.74</td><td>0.26</td><td>0.17</td></tr><tr><td>Fear</td><td>Dance Turk</td><td>0.76</td><td>0.27</td><td>0.18</td></tr><tr><td>Fear</td><td>M/S Original</td><td>0.93</td><td>0</td><td>0</td></tr><tr><td>Fear</td><td>M/S Turk</td><td>0.81</td><td>0.11</td><td>0.09</td></tr><tr><td>Joy</td><td>Dance Original</td><td>0.58</td><td>0.61</td><td>0.12</td></tr><tr><td>Joy</td><td>Dance Turk</td><td>0.48</td><td>0.71</td><td>0.1</td></tr><tr><td>Joy</td><td>M/S Original</td><td>0.53</td><td>0.73</td><td>0.11</td></tr><tr><td>Joy</td><td>M/S Turk</td><td>0.56</td><td>0.67</td><td>0.09</td></tr><tr><td colspan=\"2\">Sadness Dance Original</td><td>0.6</td><td>0.36</td><td>0.18</td></tr><tr><td>Sadness</td><td>Dance Turk</td><td>0.65</td><td>0.54</td><td>0.21</td></tr><tr><td>Sadness</td><td>M/S Original</td><td>0.66</td><td>0.36</td><td>0.03</td></tr><tr><td>Sadness</td><td>M/S Turk</td><td>0.54</td><td>0.58</td><td>0.16</td></tr><tr><td colspan=\"2\">Surprise Dance Original</td><td>0.87</td><td>0</td><td>0</td></tr><tr><td>Surprise</td><td>Dance Turk</td><td>0.86</td><td>0</td><td>0</td></tr><tr><td>Surprise</td><td>M/S Original</td><td>0.98</td><td>0</td><td>0</td></tr><tr><td>Surprise</td><td>M/S Turk</td><td>0.91</td><td>0</td><td>0</td></tr></table>",
"html": null,
"text": "Naive Bayes Trained on Augmented CBET",
"type_str": "table"
},
"TABREF18": {
"num": null,
"content": "<table><tr><td colspan=\"2\">Emotion Train</td><td>Test</td><td colspan=\"4\">Accuracy AUC Prec Rec</td></tr><tr><td>Anger</td><td>TEC</td><td>Dance</td><td>0.85</td><td colspan=\"3\">0.58 0.39 0.22</td></tr><tr><td>Anger</td><td>TEC</td><td>M/S</td><td>0.89</td><td colspan=\"3\">0.55 0.29 0.15</td></tr><tr><td>Anger</td><td colspan=\"2\">TEC Dance Turk</td><td>0.78</td><td colspan=\"3\">0.49 0.11 0.08</td></tr><tr><td>Anger</td><td>TEC</td><td>M/S Turk</td><td>0.92</td><td>0.56</td><td>1</td><td>0.13</td></tr><tr><td colspan=\"2\">Disgust TEC</td><td>Dance</td><td>0.78</td><td>0.5</td><td>0</td><td>0</td></tr><tr><td colspan=\"2\">Disgust TEC</td><td>M/S</td><td>0.97</td><td>0.5</td><td>0</td><td>0</td></tr><tr><td colspan=\"3\">Disgust TEC Dance Turk</td><td>0.95</td><td>0.5</td><td>0</td><td>0</td></tr><tr><td colspan=\"2\">Disgust TEC</td><td>M/S Turk</td><td>0.96</td><td>0.5</td><td>0</td><td>0</td></tr><tr><td>Fear</td><td>TEC</td><td>Dance</td><td>0.74</td><td colspan=\"3\">0.58 0.33 0.32</td></tr><tr><td>Fear</td><td>TEC</td><td>M/S</td><td>0.9</td><td>0.76</td><td colspan=\"2\">0.1 0.63</td></tr><tr><td>Fear</td><td colspan=\"2\">TEC Dance Turk</td><td>0.77</td><td>0.54</td><td colspan=\"2\">0.3 0.18</td></tr><tr><td>Fear</td><td>TEC</td><td>M/S Turk</td><td>0.8</td><td colspan=\"3\">0.61 0.27 0.36</td></tr><tr><td>Joy</td><td>TEC</td><td>Dance</td><td>0.67</td><td colspan=\"3\">0.66 0.65 0.54</td></tr><tr><td>Joy</td><td>TEC</td><td>M/S</td><td>0.66</td><td colspan=\"3\">0.66 0.68 0.62</td></tr><tr><td>Joy</td><td colspan=\"2\">TEC Dance Turk</td><td>0.68</td><td colspan=\"3\">0.68 0.74 0.63</td></tr><tr><td>Joy</td><td>TEC</td><td>M/S Turk</td><td>0.66</td><td colspan=\"3\">0.66 0.62 0.65</td></tr><tr><td colspan=\"2\">Sadness TEC</td><td>Dance</td><td>0.7</td><td>0.61</td><td colspan=\"2\">0.7 0.28</td></tr><tr><td colspan=\"2\">Sadness TEC</td><td>M/S</td><td>0.73</td><td>0.6</td><td colspan=\"2\">0.83 0.23</td></tr><tr><td colspan=\"3\">Sadness TEC Dance Turk</td><td>0.75</td><td colspan=\"3\">0.69 0.76 0.47</td></tr><tr><td colspan=\"2\">Sadness TEC</td><td>M/S Turk</td><td>0.51</td><td colspan=\"3\">0.49 0.46 0.13</td></tr><tr><td colspan=\"2\">Surprise TEC</td><td>Dance</td><td>0.85</td><td>0.49</td><td>0</td><td>0</td></tr><tr><td colspan=\"2\">Surprise TEC</td><td>M/S</td><td>0.97</td><td>0.49</td><td>0</td><td>0</td></tr><tr><td colspan=\"3\">Surprise TEC Dance Turk</td><td>0.88</td><td>0.5</td><td>0</td><td>0</td></tr><tr><td colspan=\"2\">Surprise TEC</td><td>M/S Turk</td><td>0.9</td><td>0.49</td><td>0</td><td>0</td></tr><tr><td/><td/><td/><td/><td/><td/><td>: Random Forest Trained on Transformed</td></tr><tr><td/><td/><td/><td/><td/><td/><td>CBET</td></tr><tr><td/><td/><td/><td/><td/><td/><td>Naive Bayes and Random Forest models, while</td></tr><tr><td/><td/><td/><td/><td/><td/><td>models for surprise were more or less equivalent</td></tr></table>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF19": {
"num": null,
"content": "<table/>",
"html": null,
"text": "",
"type_str": "table"
},
"TABREF21": {
"num": null,
"content": "<table/>",
"html": null,
"text": "",
"type_str": "table"
}
}
}
} |