File size: 137,790 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 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:28:31.471063Z"
},
"title": "Understanding and Explicitly Measuring Linguistic and Stylistic Properties of Deception via Generation and Translation",
"authors": [
{
"first": "Emily",
"middle": [],
"last": "Saldanha",
"suffix": "",
"affiliation": {},
"email": "emily.saldanha@pnnl.gov"
},
{
"first": "Aparna",
"middle": [],
"last": "Garimella",
"suffix": "",
"affiliation": {},
"email": "garimell@adobe.com"
},
{
"first": "Svitlana",
"middle": [],
"last": "Volkova",
"suffix": "",
"affiliation": {},
"email": "svitlana.volkova@pnnl.gov"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Massive digital disinformation is one of the main risks of modern society. Hundreds of models and linguistic analyses have been done to compare and contrast misleading and credible content online. However, most models do not remove the confounding factor of a topic or narrative when training, so the resulting models learn a clear topical separation for misleading versus credible content. We study the feasibility of using two strategies to disentangle the topic bias from the models to understand and explicitly measure linguistic and stylistic properties of content from misleading versus credible content. First, we develop conditional generative models to create news content that is characteristic of different credibility levels. We perform multi-dimensional evaluation of model performance on mimicking both the style and linguistic differences that distinguish news of different credibility using machine translation metrics and classification models. We show that even though generative models are able to imitate both the style and language of the original content, additional conditioning on both the news category and the topic leads to reduced performance. In a second approach, we perform deception style \"transfer\" by translating deceptive content into the style of credible content and vice versa. Extending earlier studies, we demonstrate that, when conditioned on a topic, deceptive content is shorter, less readable, more biased, and more subjective than credible content, and transferring the style from deceptive to credible content is more challenging than the opposite direction.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Massive digital disinformation is one of the main risks of modern society. Hundreds of models and linguistic analyses have been done to compare and contrast misleading and credible content online. However, most models do not remove the confounding factor of a topic or narrative when training, so the resulting models learn a clear topical separation for misleading versus credible content. We study the feasibility of using two strategies to disentangle the topic bias from the models to understand and explicitly measure linguistic and stylistic properties of content from misleading versus credible content. First, we develop conditional generative models to create news content that is characteristic of different credibility levels. We perform multi-dimensional evaluation of model performance on mimicking both the style and linguistic differences that distinguish news of different credibility using machine translation metrics and classification models. We show that even though generative models are able to imitate both the style and language of the original content, additional conditioning on both the news category and the topic leads to reduced performance. In a second approach, we perform deception style \"transfer\" by translating deceptive content into the style of credible content and vice versa. Extending earlier studies, we demonstrate that, when conditioned on a topic, deceptive content is shorter, less readable, more biased, and more subjective than credible content, and transferring the style from deceptive to credible content is more challenging than the opposite direction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "As online social media usage continues to grow, it is becoming easier to access a much wider va-riety of news sources than ever before. Around two-thirds of U.S. adults get at least some of their news from sources on social media. 1 However, with the lack of traditional fact-checking and verification processes that accompany more standard news sources, this leads to a significant potential for the spread of false, misleading, and harmful information. The growing impact of such information in the online environment has lead to increased attention, awareness, and efforts to understand and combat its spread (Wardle and Derakhshan, 2017; Ireton and Posetti, 2018) .",
"cite_spans": [
{
"start": 612,
"end": 641,
"text": "(Wardle and Derakhshan, 2017;",
"ref_id": "BIBREF39"
},
{
"start": 642,
"end": 667,
"text": "Ireton and Posetti, 2018)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In social media, a great deal of attention has been dedicated to detect and measure the spread and impact of deceptive news (Lazer et al., 2018; Vosoughi et al., 2018) . Many researchers (P\u00e9rez-Rosas and Mihalcea, 2015; Rashkin et al., 2017; Wang, 2017; Baly et al., 2018) have analyzed linguistic differences to build models that classify types of deceptive news content. There are several limitations to the use of classification models to understand differences between deceptive and trustworthy content. Firstly, such models may learn to rely on the most prominent distinguishing lexical features between news categories but may not learn to model more subtle stylistic differences. Secondly, these models may learn context-dependent features that will evolve as the topics and news events change over time.",
"cite_spans": [
{
"start": 124,
"end": 144,
"text": "(Lazer et al., 2018;",
"ref_id": "BIBREF13"
},
{
"start": 145,
"end": 167,
"text": "Vosoughi et al., 2018)",
"ref_id": "BIBREF37"
},
{
"start": 187,
"end": 219,
"text": "(P\u00e9rez-Rosas and Mihalcea, 2015;",
"ref_id": "BIBREF23"
},
{
"start": 220,
"end": 241,
"text": "Rashkin et al., 2017;",
"ref_id": "BIBREF26"
},
{
"start": 242,
"end": 253,
"text": "Wang, 2017;",
"ref_id": "BIBREF38"
},
{
"start": 254,
"end": 272,
"text": "Baly et al., 2018)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our major contribution is to understand and explicitly measure linguistic and stylistic properties of content from misleading versus credible news sources while mitigating the topical bias. For that, we develop and rigorously evaluate two separate strategies: a generation approach to generate con-tent of different credibility news source categories and a translation approach, which involves converting deceptive tweets to credible and vice versa while preserving the meaning. Our goal is to generate corresponding text in the style of credible and misleading news sources that will allow analysis of stylistic differences only, controlling for topical differences that would exist in the full corpus of real tweets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "First, for the text generation task, we aim to learn a generative language model that can produce news content characteristic of news sources of varied credibility in order to separate the linguistic and stylistic differences of these news sources. We aim to determine whether such models can reliably produce text that is characteristic of news sources of varied credibility, whether we can additionally control the topic of such generated text, and how we can best evaluate the performance of the generative models. We demonstrate that we can effectively generate text of different news source categories but that performance is reduced when additionally conditioning on topical indicators. Therefore, to perform linguistic analysis controlling for topic we turn to the translation task, for which we first create a parallel corpus and then take advantage of encoder-decoder architectures to train a transformation function to convert misleading to credible content and vice versa. Rather than aiming to modify the information content of the text, we aim to modify the stylistic properties while leaving the content as unchanged as possible.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Efforts related to the detection of deception in written text has examined deceptive language in several domains such as fake news (Conroy et al., 2015) , political speeches, online opinions about topics such as abortion or death penalty (Mihalcea and Strapparava, 2009; Newman et al., 2003) . Most of the existing models for deception detection rely on linguistic features such as n-grams, language complexity, part-of-speech tags, and syntactic and semantic features (Mihalcea and Strapparava, 2009; P\u00e9rez-Rosas and Mihalcea, 2015; Yancheva and Rudzicz, 2013) . Linguistic Inquiry and Word Count (LIWC) (Pennebaker et al., 2001) features were used to show that deceptive texts had fewer self-references, to avoid own self involved in the lies, and more words related to certainty (always, all, truly), possibly due to the need of the speaker to emphasize the \"fake\" truth (Mihalcea and Strapparava, 2009) . Recent work in deception detection focused on developing predictive models to classify fake and verified news (Conroy et al., 2015; Rubin et al., 2016) , assess information credibility (Wang, 2017), understand linguistic cues that distinguish fake and real news stories (Gravanis et al., 2019) , and characterize the signatures of coordination (Alizadeh et al., 2020; Linvill and Warren, 2020) . Recently, linguistically infused neural network architectures were developed for (a) classifying social media posts as credible or deceptive or as different types of deceptive news , (b) factuality assessments of statements with different levels of credibility (Rashkin et al., 2017) , and (c) classifying deceptive strategies and understanding intent behind deception (Volkova and Jang, 2018) .",
"cite_spans": [
{
"start": 131,
"end": 152,
"text": "(Conroy et al., 2015)",
"ref_id": "BIBREF3"
},
{
"start": 238,
"end": 270,
"text": "(Mihalcea and Strapparava, 2009;",
"ref_id": "BIBREF19"
},
{
"start": 271,
"end": 291,
"text": "Newman et al., 2003)",
"ref_id": "BIBREF20"
},
{
"start": 469,
"end": 501,
"text": "(Mihalcea and Strapparava, 2009;",
"ref_id": "BIBREF19"
},
{
"start": 502,
"end": 533,
"text": "P\u00e9rez-Rosas and Mihalcea, 2015;",
"ref_id": "BIBREF23"
},
{
"start": 534,
"end": 561,
"text": "Yancheva and Rudzicz, 2013)",
"ref_id": "BIBREF42"
},
{
"start": 605,
"end": 630,
"text": "(Pennebaker et al., 2001)",
"ref_id": "BIBREF22"
},
{
"start": 874,
"end": 906,
"text": "(Mihalcea and Strapparava, 2009)",
"ref_id": "BIBREF19"
},
{
"start": 1019,
"end": 1040,
"text": "(Conroy et al., 2015;",
"ref_id": "BIBREF3"
},
{
"start": 1041,
"end": 1060,
"text": "Rubin et al., 2016)",
"ref_id": "BIBREF29"
},
{
"start": 1179,
"end": 1202,
"text": "(Gravanis et al., 2019)",
"ref_id": "BIBREF7"
},
{
"start": 1253,
"end": 1276,
"text": "(Alizadeh et al., 2020;",
"ref_id": "BIBREF0"
},
{
"start": 1277,
"end": 1302,
"text": "Linvill and Warren, 2020)",
"ref_id": "BIBREF15"
},
{
"start": 1566,
"end": 1588,
"text": "(Rashkin et al., 2017)",
"ref_id": "BIBREF26"
},
{
"start": 1674,
"end": 1698,
"text": "(Volkova and Jang, 2018)",
"ref_id": "BIBREF35"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Text generation models leveraging deep learning architectures and neural language models have been applied to domains ranging from biographies (Lebret et al., 2016) to conversational text (Ghosh et al., 2017) , and recently to generating rumors (Ma et al., 2019) , fake news (Zellers et al., 2019) and stylometry (Schuster et al., 2020) . Large-scale pre-trained language models, such as the generative pretrained transformer models, GPT-2 (Radford et al., 2019) and GPT-3 (Brown et al., 2020) , have achieved excellent performance in freeform text generation, while approaches specifically for controllable text generation have been developed e.g., Plug and Play (Dathathri et al., 2020) and the Conditional Transformer Language (CTRL) model (Keskar et al., 2019) .",
"cite_spans": [
{
"start": 143,
"end": 164,
"text": "(Lebret et al., 2016)",
"ref_id": "BIBREF14"
},
{
"start": 188,
"end": 208,
"text": "(Ghosh et al., 2017)",
"ref_id": "BIBREF6"
},
{
"start": 245,
"end": 262,
"text": "(Ma et al., 2019)",
"ref_id": "BIBREF18"
},
{
"start": 275,
"end": 297,
"text": "(Zellers et al., 2019)",
"ref_id": "BIBREF43"
},
{
"start": 313,
"end": 336,
"text": "(Schuster et al., 2020)",
"ref_id": "BIBREF30"
},
{
"start": 440,
"end": 462,
"text": "(Radford et al., 2019)",
"ref_id": "BIBREF25"
},
{
"start": 467,
"end": 493,
"text": "GPT-3 (Brown et al., 2020)",
"ref_id": null
},
{
"start": 664,
"end": 688,
"text": "(Dathathri et al., 2020)",
"ref_id": "BIBREF4"
},
{
"start": 743,
"end": 764,
"text": "(Keskar et al., 2019)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Unlike any prior work on understanding the language of deception, we propose to disentangle the topic bias via translation and generation approaches to understand and measure stylistic and linguistic differences between misleading and credible news of different levels of credibility ranging from disinformation to credible news.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "While we would ideally like to identify deceptiveness at the level of individual news stories, due to the difficulty of this task and the high potential for misclassification, we instead focus on the classification of text at the news source level similar to (Lazer et al., 2018; Vosoughi et al., 2018) . The news sources that we study are classified into the following categories based on publicly available lists of news sources annotated by experts. \u2022 Disinformation sources provide partial, distorted, or false depictions of reality. with no intent to deceive the audience. Further details about the annotation approach can be found in . Tweets were collected from the identified Twitter accounts from for the selected news sources during the 13-month period from January 2016 and January 2017 via the public Twitter API . The distribution of tweets comprising 319K tweets from 231 deceptive news accounts, and 1M tweets from 340 credible news accounts.",
"cite_spans": [
{
"start": 259,
"end": 279,
"text": "(Lazer et al., 2018;",
"ref_id": "BIBREF13"
},
{
"start": 280,
"end": 302,
"text": "Vosoughi et al., 2018)",
"ref_id": "BIBREF37"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "3"
},
{
"text": "Topical Indicators for Generation Task We aim to develop generation models that can generate text conditional on both news source type as well as specified topical content. As a proxy for content topic, we condition on different combinations of named entities mentioned in the tweets. We detect named entities in the tweets using the AllenNLP models 2 and annotate each tweet with a one-hot encoded indicator of the presence of each of the most common 250 named entities in the corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "3"
},
{
"text": "Parallel Corpus Creation To support the translation task, we aligned tweet pairs from credible and deceptive news sources when they both were published on the same day (temporal proximity) and when the pair meets our threshold of content similarity. That is, when (a) all named entities and at least 70% (manually determined threshold) of the nouns in the deceptive tweet also occurs in the credible tweet, and (b) the (subject, verb, object) tuples obtained using SyntaxNet dependency parser in both deceptive and credible tweets match.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "3"
},
{
"text": "However, we do not want to have pairs with perfect similarities, as it would not serve our goal to study the differences between the style of credible and deceptive news sources. To obtain the final set of tweet pairs from the aligned set, we compute the text similarities for each pair using edit distance and cosine distances of Word2Vec, Doc2Vec, and TFIDF. After rigorous manual inspection we retain only those pairs with edit similarity that falls within the (0.35, 0.7) range. This results in 105,365 aligned (credible, deceptive) tweet pairs. 3",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "3"
},
{
"text": "Generation Models To support our feasibility analysis, we leverage a three layer Long Short-Term Memory (LSTM) word-level language model for text generation. To create the conditional generative models, we add an embedding layer for the news source category that learns a dense representation for each of the six categories. The relevant embedding is appended to each token embedding in the input sequence, such that the model is tasked with predicting the next token given a modified version of the input token representation that is specific to the class of the input. When conditioning on named entity in addition to the news source, we add an additional component to the embedding by applying a dense layer to a one hot encoded representation of 250 most frequent named entities present in the tweet. For all models we use the Adam optimizer, 256-dimensional hidden layers, a batch size of 256, and 200 epochs trained on a Tesla P40 GPU.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "4"
},
{
"text": "Translation Models Sequence-to-sequence neural models have significantly advanced the state-ofthe-art in a variety of natural language processing tasks such as machine translation, speech recognition, and text summarization (Sutskever et al., 2014) . We propose multiple variations of encoderdecoder models that learn a function that transforms tweets from deceptive news sources to the style of credible news sources, and the other way around. We run experiments with two commonlyused types of recurrent layers -LSTMs and Gated Recurrent Units (GRUs) -for the encoders and decoders. For both LSTMs and GRUs we used the Adam optimizer with a learning rate of 0.001, 256-dimensional embeddings, categorical crossentropy loss, a batch size of 64, and 100 epochs with early stopping. Given the success of attention mechanisms for language tasks, we additionally experiment with multiplicative- (Luong et al., 2015) and self-attention (Vaswani et al., 2017 ) mechanisms.We conduct translations in both directions:",
"cite_spans": [
{
"start": 224,
"end": 248,
"text": "(Sutskever et al., 2014)",
"ref_id": "BIBREF33"
},
{
"start": 891,
"end": 911,
"text": "(Luong et al., 2015)",
"ref_id": "BIBREF17"
},
{
"start": 931,
"end": 952,
"text": "(Vaswani et al., 2017",
"ref_id": "BIBREF34"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "4"
},
{
"text": "(1) forward: credible to deceptive, and (2) reverse: deceptive to credible.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "4"
},
{
"text": "A conceptual schematic of both the generation and translation approaches is shown in Figure 1 .",
"cite_spans": [],
"ref_spans": [
{
"start": 85,
"end": 93,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Approach",
"sec_num": "4"
},
{
"text": "For the generation task, we aim to evaluate both the quality of the generated text and the similarity of the generated text to the desired conditional news category. The sign of good model performance is when the text similarity is higher between generated tweets of a specific category and real tweets of the same category than it is with real tweets of a different category.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Results",
"sec_num": "5"
},
{
"text": "We first evaluate the performance of the models that were conditioned on only the news source category. Examples of tweets of each category that were generated using these models can be found in Table 1 . We next evaluate how the addition of topical conditioning affects the performance of the generative models. In particular, we are interested in how the additional conditioning affects the ability of the model to continue to generate tweets that are characteristic of the news source categories. Overall, we find that model performance on mimicking the differences between the news source types decreases due to an increasing complexity of the task with additional conditioning.",
"cite_spans": [],
"ref_spans": [
{
"start": 195,
"end": 202,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generation Results",
"sec_num": "5"
},
{
"text": "Generation Evaluation with BLEU In order to determine the textual similarity of tweets both within and across categories, we calculate the BLEU scores of both real and generated tweets from a given news source category with the same category and with different categories. Because we do not have one-to-one correspondence between real and generated text examples for the generation task, we calculate the BLEU score of each generated example against a sample of the full real corpus. We equivalently generate a BLEU score for a sample of real tweets against the real corpus to determine the typical n-gram similarity of real tweets to other real tweets. By comparing these two scores, we can determine how the similarity of generated samples compares with the expected similarity of real samples. Figure 2 summarizes the BLEU score results for both real and generated content. We find that the BLEU score similarity to the target category is similar to what is observed for real content from the same category. However, the BLEU score similarity to content from categories other than the target category is significantly higher than what is observed for real cross-category content. For all categories we find that the BLEU score ratio of the generated content is greater than one, indicating that generated content is more similar to the target category than the other categories. However, in all cases the ratio is still significantly less than what is observed for the real content indicating that the models may be \"hedging\" and skewing their results towards similarity to the full corpus. We find that generated content that are characteristic of disinformation news sources come the closest to matching the BLEU-score ratio of real tweets.",
"cite_spans": [],
"ref_spans": [
{
"start": 797,
"end": 805,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generation Results",
"sec_num": "5"
},
{
"text": "For models which were conditioned on named entities, we can see that the raw BLEU scores are lower both within category and across category compared with the simpler model. This indicates that this model produces text that differs more from the real tweet corpus as a whole. Additionally, the BLEU score ratio has been reduced to below 1.0 for the hoax news category, indicating that these generated tweets are not more similar to real hoax tweets than real tweets of other categories. These results might be explained by the additional conditioning causing the generated distribution to differ from the real distribution. Because our goal is to NEWS TYPE GENERATED TWEETS Clickbait 1) \" the congress is a dangerous guy . \" 2) this is the reason why the ' [OOV] ' was the most popular isis group Verified 1) punjab police arrest two suspects suspected of carrying explosives 2) # breaking : cnn projects [@] wins florida , democratic votes Table 1 : Example content generated for each news source type category using news source type conditional model. Figure 2 : The BLEU scores of generated text (orange and green) calculated relative to real tweets of the target category (left) and real tweets of the other categories (middle), as well as the ratio of the BLEU score with the target category to the BLEU score with other categories (right). As a benchmark, the metrics for the generated text are compared with the BLEU scores of a sample of real tweets with other real tweets (blue).",
"cite_spans": [],
"ref_spans": [
{
"start": 940,
"end": 947,
"text": "Table 1",
"ref_id": null
},
{
"start": 1053,
"end": 1061,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generation Results",
"sec_num": "5"
},
{
"text": "generate text that stylistically matches the different news categories, even though we force the model to focus on topic areas that are not typical of that category, we aim to probe the stylistic similarity of real and generated text. Below we explore additional dimensions of evaluation to probe the stylistic quality of the conditional generation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Results",
"sec_num": "5"
},
{
"text": "To probe the ability of the conditional model to mimic both content and stylistic differences among the news categories, we train models to predict news source category on both real and generated tweets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "Note that the focus of this work is not to train the best deception detection model. Instead, it is to use the classification models to evaluate and contrast the performance of the generation models.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "We evaluate each trained model on both real and generated tweets to determine how well the distinguishing patterns learned on each data set transfers to the other. We compare the performance across several different train and test splits. We train the classification model on the same tweets used to train the generative model and evaluate its perfor-mance on a validation set of real tweets and on the generated tweets. We also train the classification models on the validation set of real tweets and evaluate on the generated tweets. Finally, we train the classification model on the generated tweets and evaluate the performance on the real tweets from the generative training data, the real data from the validation set, and a set of held-out validation generated tweets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "To understand whether the generative models are successfully mimicking both the stylistic and linguistic differences of the news categories, we train models on several variations of the input text. First, we compare feed-forward models with bagof-words (BoW) inputs to sequence-based LSTM models to compare the performance when given just topical word information versus stylistic phrasing information. Secondly, we compare using the true content words of the tweet with a processed version in which the content words are replaced with placeholders leaving only punctuation and other stylistic markers intact. If the classification models generalize well between the real and gener- Table 2 : Classification model performance for evaluation of the generation task, including performance for news source and named entity conditional generated tweets for different train and test sets used for the classification model and different data representations for the text. Performance on this classification task illustrates whether the generated texts have the same discriminative content and stylistic features as the real text.",
"cite_spans": [],
"ref_spans": [
{
"start": 683,
"end": 690,
"text": "Table 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "ated text using just the stylistic information, we can infer that the generative models are able to model the tweet style in addition to the more salient content differences. Table 2 summarizes the results of model evaluation via classification on the tweets. For the news source type conditional model, we find that the models trained on the real tweets are able to learn distinguishing features that generalize to the generated tweets, including both when topical information is included or excluded. This indicates that the generated content is replicating many of the features that characterized the differences among news source types. However, models trained on the generated tweets do not generalize well to the real text. This indicates that, while the generated tweets replicate many of the features that distinguish tweets of different credibility, they likely also include \"cheats\" that are not present in the real tweets.",
"cite_spans": [],
"ref_spans": [
{
"start": 175,
"end": 182,
"text": "Table 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "We find that removing the content information reduces model performance. However, similar reduction in performance is observed when testing on both real and generated tweets when training on real data. This indicates that the generated tweets are retaining both content and style features that distinguish the different news categories. Interestingly, the performance of models trained on the generated tweets does not drop as much when the content is removed, indicating that for the generated tweets the structural characteristics are more distinct among categories.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "The classification performance on distinguishing the fine-grained credibility categories of the generated text sharply declines when conditioning on both news source type and the named entities. This indicates that the generated text is no longer replicating the same distinguishing features of the real text. Interestingly, this is also true for the classification models that leverage only stylistic information and not content information, indicating that the topical conditioning shifts the stylistic features of the text in addition to the content features.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Classification",
"sec_num": null
},
{
"text": "As a final evaluation dimension, we compare the linguistic properties of generated text with real text using both readability measures and LIWC features. Because we are aiming to replicate the differences among text categories in the generated text, we evaluate the relative values of the linguistic measures among the news source categories. We first calculate the mean and standard deviation of each measure for each news source category of both generated and real text. Then we rank the different categories in terms of both their mean value and standard deviation and compare the observed rankings of the generated text with the observed ranking of the real text. This comparison is performed using the Spearman correlation. The comparison of the rankings of the mean value reveals whether the generated text follows the same relative linguistic patterns as the real text, while the comparison of the ranking of the standard deviation tells us whether the generated text shows the same pattern of variability across categories as the real text. Figure 3 : Spearman correlation between the ranking of news categories of real tweets and generated tweets for the mean (a) and standard deviation (b) for different types of readability and LIWC measures. High correlation for the upper plot would mean that the generated text correctly replicated which categories has higher values of the linguistic features than others, while high correlation in the lower plots would mean that the generated text correctly replicated which categories had higher variability in linguistic properties than others. Figure 3 presents the comparison of the relative linguistic feature values between the real and generated tweets. The generated text replicated the observed patterns in the readability statistics among categories very well, with an average Spearman correlation of the mean readability metrics of 0.98 and the average correlation of the standard deviation of these metrics of 0.84.",
"cite_spans": [],
"ref_spans": [
{
"start": 1049,
"end": 1057,
"text": "Figure 3",
"ref_id": null
},
{
"start": 1597,
"end": 1605,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Generation Evaluation via Linguistic Analysis",
"sec_num": null
},
{
"text": "The average Spearman correlation of the mean LIWC features values among categories was 0.71 and the average correlation of the standard deviation of the features was 0.53. We find that on average all types of LIWC features have positive correlation, but that the cognitive type features, e.g. causality and certainty, time and space features, e.g. motion and directions, and the linguistic type features, e.g. pronoun usage and negation, have the strongest agreement with the real text. The LIWC features with low correlation between the generated and real tweets tend to be those features which leverage keywords that are relatively rare in the true data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Linguistic Analysis",
"sec_num": null
},
{
"text": "In comparison with the simpler conditional models, we find a lower average Spearman correlation for both readability and LIWC measures for the named entity conditional model. For the readability measures, the correlation of the mean values was 0.70 and the correlation of the standard deviations was 0.76. For the LIWC features, the correlation of the mean values was 0.41 and the correlation of the standard deviations was 0.35.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Evaluation via Linguistic Analysis",
"sec_num": null
},
{
"text": "We have leveraged three different evaluation approaches to probe the ability of the generation models to mimic both the language and the style of different news categories. These metrics have consistently shown that the news source category conditional models are able to successfully replicate distinguishing features of the different categories in terms of both content and style. However, this ability, in terms of both style and content, is significantly reduced when conditioning on additional topical information in the form of named entities. Because we do not achieve sufficient performance of the topic conditional models, further development is needed before these models can be leveraged to control for topic bias in analysis of the stylistic differences between credible and deceptive news sources.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generation Results Summary",
"sec_num": null
},
{
"text": "To evaluate the translation task, we rely on a set of metrics used to evaluate machine translation and text generation systems: BLEU (Papineni et al., 2002) , iBLEU (Sun and Zhou, 2012) , FK-BLEU (Xu et al., 2016) , SARI (Xu et al., 2016) , and GLEU (Wu et al., 2016) . We present performance results of translations between deceptive and credible aligned tweets in Table 3 . All encoderdecoder models that learn transformations from credible into deceptive (forward translation) are more accurate than those that learn translations from deceptive into credible (reverse translation). GRUs perform significantly better than LSTMs, while GRUs perform even more accurately when augmented with the attention mechanism. The addition of self-attention results in the highest trans- Table 4 : Examples of translations from misleading news sources to credible using our best model with the selfattention.",
"cite_spans": [
{
"start": 133,
"end": 156,
"text": "(Papineni et al., 2002)",
"ref_id": "BIBREF21"
},
{
"start": 165,
"end": 185,
"text": "(Sun and Zhou, 2012)",
"ref_id": "BIBREF32"
},
{
"start": 196,
"end": 213,
"text": "(Xu et al., 2016)",
"ref_id": "BIBREF41"
},
{
"start": 221,
"end": 238,
"text": "(Xu et al., 2016)",
"ref_id": "BIBREF41"
},
{
"start": 250,
"end": 267,
"text": "(Wu et al., 2016)",
"ref_id": "BIBREF40"
}
],
"ref_spans": [
{
"start": 366,
"end": 373,
"text": "Table 3",
"ref_id": "TABREF3"
},
{
"start": 777,
"end": 784,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Translation Results",
"sec_num": "6"
},
{
"text": "lation performance in both the forward and the reverse directions. To qualitatively demonstrate model performance, we present example outputs including good and bad translations from our best model in Table 4 .",
"cite_spans": [],
"ref_spans": [
{
"start": 201,
"end": 208,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Translation Results",
"sec_num": "6"
},
{
"text": "We leverage our learned translation models to examine the stylistic properties of credible versus misleading news sources independent of the topic. We perform pairwise comparative linguistic analysis of real tweets from credible news sources in comparison with their corresponding translated deceptive counterparts using readability, biased and subjective language measures. We also compare stylistic features including the average number of edit operations (insertions, deletions, and substitutions) required to transform a tweet from a credible news source into one from a deceptive one, and vice versa, and differences in tweet length. We report the results for each measure in Table 5 . We find that tweets from credible news sources are longer, on average, with an average length of 73 characters compared to 68 for misleading content. Consistently, we find that an average of 6 character deletions are needed to go from credible to deceptive and 6 insertions to transform vice versa.",
"cite_spans": [],
"ref_spans": [
{
"start": 681,
"end": 688,
"text": "Table 5",
"ref_id": "TABREF5"
}
],
"eq_spans": [],
"section": "Topic-Controlled Linguistic and Stylistic Analysis",
"sec_num": "6.1"
},
{
"text": "To compare readability of content from credible and misleading news sources conditioned on the topic of the post, we apply several widely-used readability measures: Automated Readability Index (ARI) (Senter and Smith, 1967) , Flesch Reading Ease (FRE) (Farr et al., 1951) , Flesch-Kincaid Grade Level (FKGL) (Kincaid et al., 1975) , the number of syllables, and the number of complex words. A given tweet is more readable when its ARI and FKGL are low, but FRE is high. On an average, narratives from credible news sources are more readable compared to narratives from misleading news sources -i.e., they have lower ARI and FKG, higher FRE scores, have more syllables, and fewer complex words. We use publicly available subjectivity lexicons (Riloff and Wiebe, 2003; Liu et al., 2005) to annotate strongly negative, weakly negative, weakly positive, and strongly positive terms in the tweets. We contrast the average fraction of terms in each tweet that belong to one of the four groups in tweets from credible and deceptive news sources. In line with earlier work, we confirm that content from deceptive sources has more subjective terms, even when conditioned on topic.",
"cite_spans": [
{
"start": 199,
"end": 223,
"text": "(Senter and Smith, 1967)",
"ref_id": "BIBREF31"
},
{
"start": 252,
"end": 271,
"text": "(Farr et al., 1951)",
"ref_id": "BIBREF5"
},
{
"start": 308,
"end": 330,
"text": "(Kincaid et al., 1975)",
"ref_id": "BIBREF12"
},
{
"start": 742,
"end": 766,
"text": "(Riloff and Wiebe, 2003;",
"ref_id": "BIBREF28"
},
{
"start": 767,
"end": 784,
"text": "Liu et al., 2005)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Topic-Controlled Linguistic and Stylistic Analysis",
"sec_num": "6.1"
},
{
"text": "We compute the average fraction of terms in each tweet that fall under one of the following verb types: assertive verbs (Hooper, 1974) that bring emphasis to a sentence (point out, claim), implicative verbs (Karttunen, 1971) whose factuality depends upon a condition (avoid, hesitate), and report verbs (Recasens et al., 2013) that indicate that discourse is being quoted or paraphrased (admit, criticize). We found that controlling for topic, credible content has a higher fraction of assertive and report verbs, while implicative verbs occur more often in deceptive tweets.",
"cite_spans": [
{
"start": 120,
"end": 134,
"text": "(Hooper, 1974)",
"ref_id": "BIBREF8"
},
{
"start": 207,
"end": 224,
"text": "(Karttunen, 1971)",
"ref_id": "BIBREF10"
},
{
"start": 303,
"end": 326,
"text": "(Recasens et al., 2013)",
"ref_id": "BIBREF27"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Topic-Controlled Linguistic and Stylistic Analysis",
"sec_num": "6.1"
},
{
"text": "If we compare these results to existing results which directly compare tweets from credible and deceptive sources without controlling for topic, we find that some conclusions are confirmed by this topic-controlled analysis while others show a discrepancy. Consistent with and (Rashkin et al., 2017) , we find that deceptive news sources use more subjective and biased language than credible sources. However, in contrast with the earlier studies, we find that after controlling for topic credible news sources are more likely to use assertive and report verbs.",
"cite_spans": [
{
"start": 276,
"end": 298,
"text": "(Rashkin et al., 2017)",
"ref_id": "BIBREF26"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Topic-Controlled Linguistic and Stylistic Analysis",
"sec_num": "6.1"
},
{
"text": "We have presented a novel understanding of linguistic and stylistic properties deception using translation and generation approaches designed to disentangle from the topic bias.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future Work",
"sec_num": "7"
},
{
"text": "We have demonstrated progress towards developing generative models for this purpose, with our generative models being able to imitate both the style and the content of the real tweets. We have evaluated our generative models using BLEU scores, linguistic analysis, and classification. However, with additional topical conditioning dimensions, we find significantly reduced performance on maintaining the observed stylistic and content differences. In order to improve the performance of these generative models, future work includes the application of more advanced controllable text generation model transformer-based and other recently emerged neural architectures (Brown et al., 2020; Prabhumoye et al., 2020; Keskar et al., 2019; Dathathri et al., 2020; Radford et al., 2019) . We will focus on improving the ability to perform multi-dimensional conditioning to enable the desired topic-controlled misleading versus credible news source style analysis.",
"cite_spans": [
{
"start": 667,
"end": 687,
"text": "(Brown et al., 2020;",
"ref_id": "BIBREF2"
},
{
"start": 688,
"end": 712,
"text": "Prabhumoye et al., 2020;",
"ref_id": "BIBREF24"
},
{
"start": 713,
"end": 733,
"text": "Keskar et al., 2019;",
"ref_id": "BIBREF11"
},
{
"start": 734,
"end": 757,
"text": "Dathathri et al., 2020;",
"ref_id": "BIBREF4"
},
{
"start": 758,
"end": 779,
"text": "Radford et al., 2019)",
"ref_id": "BIBREF25"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future Work",
"sec_num": "7"
},
{
"text": "We have shown that translating from misleading to credible content is more difficult than the opposite direction. This may be because deceptive content has a higher level of stylistic and lexical variation making it more difficult for the models correctly anticipate its style, while verified content has a more factual style. Our translation results clearly demonstrate that, when conditioned on a topic, content from deceptive news sources is shorter, less readable, more biased, and more subjective than content from deceptive news sources. Leveraging these models to allow topic-agnostic comparison of the style of deceptive and credible news sources, we demonstrate several key stylistic differences.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future Work",
"sec_num": "7"
},
{
"text": "https://www.journalism.org/2018/09/10/news-useacross-social-media-platforms-2018/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://demo.allennlp.org/named-entity-recognition",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The tweet data set, parallel corpus, and trained models will be made publicly available upon acceptance.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Content-based features predict social media influence operations",
"authors": [
{
"first": "Meysam",
"middle": [],
"last": "Alizadeh",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Jacob",
"suffix": ""
},
{
"first": "Cody",
"middle": [],
"last": "Shapiro",
"suffix": ""
},
{
"first": "Joshua",
"middle": [
"A"
],
"last": "Buntain",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Tucker",
"suffix": ""
}
],
"year": 2020,
"venue": "Science advances",
"volume": "6",
"issue": "30",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Meysam Alizadeh, Jacob N Shapiro, Cody Buntain, and Joshua A Tucker. 2020. Content-based features predict social media influence operations. Science advances, 6(30):eabb5824.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Predicting factuality of reporting and bias of news media sources",
"authors": [
{
"first": "Ramy",
"middle": [],
"last": "Baly",
"suffix": ""
},
{
"first": "Georgi",
"middle": [],
"last": "Karadzhov",
"suffix": ""
},
{
"first": "Dimitar",
"middle": [],
"last": "Alexandrov",
"suffix": ""
},
{
"first": "James",
"middle": [],
"last": "Glass",
"suffix": ""
},
{
"first": "Preslav",
"middle": [],
"last": "Nakov",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "3528--3539",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ramy Baly, Georgi Karadzhov, Dimitar Alexandrov, James Glass, and Preslav Nakov. 2018. Predict- ing factuality of reporting and bias of news media sources. In Proceedings of Empirical Methods in Natural Language Processing, pages 3528-3539.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Language models are few-shot learners",
"authors": [
{
"first": "Benjamin",
"middle": [],
"last": "Tom B Brown",
"suffix": ""
},
{
"first": "Nick",
"middle": [],
"last": "Mann",
"suffix": ""
},
{
"first": "Melanie",
"middle": [],
"last": "Ryder",
"suffix": ""
},
{
"first": "Jared",
"middle": [],
"last": "Subbiah",
"suffix": ""
},
{
"first": "Prafulla",
"middle": [],
"last": "Kaplan",
"suffix": ""
},
{
"first": "Arvind",
"middle": [],
"last": "Dhariwal",
"suffix": ""
},
{
"first": "Pranav",
"middle": [],
"last": "Neelakantan",
"suffix": ""
},
{
"first": "Girish",
"middle": [],
"last": "Shyam",
"suffix": ""
},
{
"first": "Amanda",
"middle": [],
"last": "Sastry",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Askell",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2005.14165"
]
},
"num": null,
"urls": [],
"raw_text": "Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. arXiv preprint arXiv:2005.14165.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Automatic deception detection: Methods for finding fake news",
"authors": [
{
"first": "K",
"middle": [],
"last": "Nadia",
"suffix": ""
},
{
"first": "Victoria",
"middle": [
"L"
],
"last": "Conroy",
"suffix": ""
},
{
"first": "Yimin",
"middle": [],
"last": "Rubin",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Chen",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the Association for Information Science and Technology",
"volume": "52",
"issue": "",
"pages": "1--4",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nadia K Conroy, Victoria L Rubin, and Yimin Chen. 2015. Automatic deception detection: Methods for finding fake news. Proceedings of the Association for Information Science and Technology, 52(1):1-4.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Plug and play language models: a simple approach to controlled text generation",
"authors": [
{
"first": "Sumanth",
"middle": [],
"last": "Dathathri",
"suffix": ""
},
{
"first": "Andrea",
"middle": [],
"last": "Madotto",
"suffix": ""
},
{
"first": "Janice",
"middle": [],
"last": "Lan",
"suffix": ""
},
{
"first": "Jane",
"middle": [],
"last": "Hung",
"suffix": ""
},
{
"first": "Eric",
"middle": [],
"last": "Frank",
"suffix": ""
},
{
"first": "Piero",
"middle": [],
"last": "Molino",
"suffix": ""
},
{
"first": "Jason",
"middle": [],
"last": "Yosinski",
"suffix": ""
},
{
"first": "Rosanne",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2020,
"venue": "International Conference on Learning Representations",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung, Eric Frank, Piero Molino, Jason Yosinski, and Rosanne Liu. 2020. Plug and play language mod- els: a simple approach to controlled text generation. International Conference on Learning Representa- tions.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Simplification of Flesch Reading Ease Formula",
"authors": [
{
"first": "N",
"middle": [],
"last": "James",
"suffix": ""
},
{
"first": "James",
"middle": [
"J"
],
"last": "Farr",
"suffix": ""
},
{
"first": "Donald G",
"middle": [],
"last": "Jenkins",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Paterson",
"suffix": ""
}
],
"year": 1951,
"venue": "Journal of Applied Psychology",
"volume": "35",
"issue": "5",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "James N Farr, James J Jenkins, and Donald G Paterson. 1951. Simplification of Flesch Reading Ease For- mula. Journal of Applied Psychology, 35(5):333.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Affect-LM: A neural language model for customizable affective text generation",
"authors": [
{
"first": "Sayan",
"middle": [],
"last": "Ghosh",
"suffix": ""
},
{
"first": "Mathieu",
"middle": [],
"last": "Chollet",
"suffix": ""
},
{
"first": "Eugene",
"middle": [],
"last": "Laksana",
"suffix": ""
},
{
"first": "Louis-Philippe",
"middle": [],
"last": "Morency",
"suffix": ""
},
{
"first": "Stefan",
"middle": [],
"last": "Scherer",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "634--642",
"other_ids": {
"DOI": [
"10.18653/v1/P17-1059"
]
},
"num": null,
"urls": [],
"raw_text": "Sayan Ghosh, Mathieu Chollet, Eugene Laksana, Louis-Philippe Morency, and Stefan Scherer. 2017. Affect-LM: A neural language model for customiz- able affective text generation. In Proceedings of the 55th Annual Meeting of the Association for Compu- tational Linguistics (Volume 1: Long Papers), pages 634-642, Vancouver, Canada. Association for Com- putational Linguistics.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Behind the cues: A benchmarking study for fake news detection",
"authors": [
{
"first": "Georgios",
"middle": [],
"last": "Gravanis",
"suffix": ""
},
{
"first": "Athena",
"middle": [],
"last": "Vakali",
"suffix": ""
}
],
"year": 2019,
"venue": "Expert Systems with Applications",
"volume": "128",
"issue": "",
"pages": "201--213",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Georgios Gravanis, Athena Vakali, Konstantinos Dia- mantaras, and Panagiotis Karadais. 2019. Behind the cues: A benchmarking study for fake news de- tection. Expert Systems with Applications, 128:201- 213.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "On assertive predicates",
"authors": [
{
"first": "B",
"middle": [],
"last": "Joan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Hooper",
"suffix": ""
}
],
"year": 1974,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joan B Hooper. 1974. On assertive predicates. Indiana University Linguistics Club.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Journalism, fake news & disinformation: handbook for journalism education and training",
"authors": [
{
"first": "Cherilyn",
"middle": [],
"last": "Ireton",
"suffix": ""
},
{
"first": "Julie",
"middle": [],
"last": "Posetti",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cherilyn Ireton and Julie Posetti. 2018. Journalism, fake news & disinformation: handbook for journal- ism education and training. UNESCO Publishing.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Implicative verbs. Language",
"authors": [
{
"first": "Lauri",
"middle": [],
"last": "Karttunen",
"suffix": ""
}
],
"year": 1971,
"venue": "",
"volume": "",
"issue": "",
"pages": "340--358",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lauri Karttunen. 1971. Implicative verbs. Language, pages 340-358.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Ctrl: A conditional transformer language model for controllable generation",
"authors": [
{
"first": "Bryan",
"middle": [],
"last": "Nitish Shirish Keskar",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Mccann",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Lav",
"suffix": ""
},
{
"first": "Caiming",
"middle": [],
"last": "Varshney",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Xiong",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Socher",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1909.05858"
]
},
"num": null,
"urls": [],
"raw_text": "Nitish Shirish Keskar, Bryan McCann, Lav R Varshney, Caiming Xiong, and Richard Socher. 2019. Ctrl: A conditional transformer language model for control- lable generation. arXiv preprint arXiv:1909.05858.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel",
"authors": [
{
"first": "Robert P Fishburne",
"middle": [],
"last": "Peter Kincaid",
"suffix": ""
},
{
"first": "Richard",
"middle": [
"L"
],
"last": "Jr",
"suffix": ""
},
{
"first": "Brad",
"middle": [
"S"
],
"last": "Rogers",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Chissom",
"suffix": ""
}
],
"year": 1975,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J Peter Kincaid, Robert P Fishburne Jr, Richard L Rogers, and Brad S Chissom. 1975. Derivation of new readability formulas (automated readability in- dex, fog count and flesch reading ease formula) for navy enlisted personnel.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "The science of fake news",
"authors": [
{
"first": "M",
"middle": [
"J"
],
"last": "David",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Lazer",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Matthew",
"suffix": ""
},
{
"first": "Yochai",
"middle": [],
"last": "Baum",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Benkler",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Adam",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Berinsky",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Kelly",
"suffix": ""
},
{
"first": "Filippo",
"middle": [],
"last": "Greenhill",
"suffix": ""
},
{
"first": "Miriam",
"middle": [
"J"
],
"last": "Menczer",
"suffix": ""
},
{
"first": "Brendan",
"middle": [],
"last": "Metzger",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Nyhan",
"suffix": ""
}
],
"year": 2018,
"venue": "Science",
"volume": "359",
"issue": "6380",
"pages": "1094--1096",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David MJ Lazer, Matthew A Baum, Yochai Ben- kler, Adam J Berinsky, Kelly M Greenhill, Filippo Menczer, Miriam J Metzger, Brendan Nyhan, Gor- don Pennycook, David Rothschild, et al. 2018. The science of fake news. Science, 359(6380):1094- 1096.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Neural text generation from structured data with application to the biography domain",
"authors": [
{
"first": "R\u00e9mi",
"middle": [],
"last": "Lebret",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Grangier",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Auli",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1203--1213",
"other_ids": {
"DOI": [
"10.18653/v1/D16-1128"
]
},
"num": null,
"urls": [],
"raw_text": "R\u00e9mi Lebret, David Grangier, and Michael Auli. 2016. Neural text generation from structured data with application to the biography domain. In Proceed- ings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1203-1213, Austin, Texas. Association for Computational Lin- guistics.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Troll factories: Manufacturing specialized disinformation on twitter",
"authors": [
{
"first": "L",
"middle": [],
"last": "Darren",
"suffix": ""
},
{
"first": "Patrick L",
"middle": [],
"last": "Linvill",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Warren",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "1--21",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Darren L Linvill and Patrick L Warren. 2020. Troll factories: Manufacturing specialized disinformation on twitter. Political Communication, pages 1-21.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Opinion observer: analyzing and comparing opinions on the web",
"authors": [
{
"first": "Bing",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Minqing",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "Junsheng",
"middle": [],
"last": "Cheng",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the 14th international conference on World Wide Web",
"volume": "",
"issue": "",
"pages": "342--351",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bing Liu, Minqing Hu, and Junsheng Cheng. 2005. Opinion observer: analyzing and comparing opin- ions on the web. In Proceedings of the 14th inter- national conference on World Wide Web, pages 342- 351.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Effective approaches to attention-based neural machine translation",
"authors": [
{
"first": "Thang",
"middle": [],
"last": "Luong",
"suffix": ""
},
{
"first": "Hieu",
"middle": [],
"last": "Pham",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Manning",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1412--1421",
"other_ids": {
"DOI": [
"10.18653/v1/D15-1166"
]
},
"num": null,
"urls": [],
"raw_text": "Thang Luong, Hieu Pham, and Christopher D. Man- ning. 2015. Effective approaches to attention-based neural machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natu- ral Language Processing, pages 1412-1421, Lis- bon, Portugal. Association for Computational Lin- guistics.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Detect rumors on Twitter by promoting information campaigns with generative adversarial learning",
"authors": [
{
"first": "Jing",
"middle": [],
"last": "Ma",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "Gao",
"suffix": ""
},
{
"first": "Kam-Fai",
"middle": [],
"last": "Wong",
"suffix": ""
}
],
"year": 2019,
"venue": "The World Wide Web Conference",
"volume": "",
"issue": "",
"pages": "3049--3055",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jing Ma, Wei Gao, and Kam-Fai Wong. 2019. Detect rumors on Twitter by promoting information cam- paigns with generative adversarial learning. In The World Wide Web Conference, pages 3049-3055.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "The lie detector: Explorations in the automatic recognition of deceptive language",
"authors": [
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "Carlo",
"middle": [],
"last": "Strapparava",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the ACL-IJCNLP 2009 Conference Short Papers",
"volume": "",
"issue": "",
"pages": "309--312",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rada Mihalcea and Carlo Strapparava. 2009. The lie detector: Explorations in the automatic recognition of deceptive language. In Proceedings of the ACL- IJCNLP 2009 Conference Short Papers, pages 309- 312. Association for Computational Linguistics.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Lying words: Predicting deception from linguistic styles. Personality and social psychology bulletin",
"authors": [
{
"first": "James",
"middle": [
"W"
],
"last": "Matthew L Newman",
"suffix": ""
},
{
"first": "Diane",
"middle": [
"S"
],
"last": "Pennebaker",
"suffix": ""
},
{
"first": "Jane",
"middle": [
"M"
],
"last": "Berry",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Richards",
"suffix": ""
}
],
"year": 2003,
"venue": "",
"volume": "29",
"issue": "",
"pages": "665--675",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matthew L Newman, James W Pennebaker, Diane S Berry, and Jane M Richards. 2003. Lying words: Predicting deception from linguistic styles. Person- ality and social psychology bulletin, 29(5):665-675.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "BLEU: A method for automatic evaluation of machine translation",
"authors": [
{
"first": "Kishore",
"middle": [],
"last": "Papineni",
"suffix": ""
},
{
"first": "Salim",
"middle": [],
"last": "Roukos",
"suffix": ""
},
{
"first": "Todd",
"middle": [],
"last": "Ward",
"suffix": ""
},
{
"first": "Wei-Jing",
"middle": [],
"last": "Zhu",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, ACL '02",
"volume": "",
"issue": "",
"pages": "311--318",
"other_ids": {
"DOI": [
"10.3115/1073083.1073135"
]
},
"num": null,
"urls": [],
"raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. BLEU: A method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting on Association for Compu- tational Linguistics, ACL '02, page 311-318, USA. Association for Computational Linguistics.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Linguistic inquiry and word count: LIWC",
"authors": [
{
"first": "Martha",
"middle": [
"E"
],
"last": "James W Pennebaker",
"suffix": ""
},
{
"first": "Roger J",
"middle": [],
"last": "Francis",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Booth",
"suffix": ""
}
],
"year": 2001,
"venue": "Mahway: Lawrence Erlbaum Associates",
"volume": "71",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "James W Pennebaker, Martha E Francis, and Roger J Booth. 2001. Linguistic inquiry and word count: LIWC 2001. Mahway: Lawrence Erlbaum Asso- ciates, 71(2001):2001.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Experiments in open domain deception detection",
"authors": [
{
"first": "Ver\u00f3nica",
"middle": [],
"last": "P\u00e9rez",
"suffix": ""
},
{
"first": "-",
"middle": [],
"last": "Rosas",
"suffix": ""
},
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1120--1125",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ver\u00f3nica P\u00e9rez-Rosas and Rada Mihalcea. 2015. Ex- periments in open domain deception detection. In Proceedings of Empirical Methods in Natural Lan- guage Processing, pages 1120-1125.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Exploring controllable text generation techniques",
"authors": [
{
"first": "Shrimai",
"middle": [],
"last": "Prabhumoye",
"suffix": ""
},
{
"first": "Alan",
"middle": [
"W"
],
"last": "Black",
"suffix": ""
},
{
"first": "Ruslan",
"middle": [],
"last": "Salakhutdinov",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2005.01822"
]
},
"num": null,
"urls": [],
"raw_text": "Shrimai Prabhumoye, Alan W Black, and Rus- lan Salakhutdinov. 2020. Exploring control- lable text generation techniques. arXiv preprint arXiv:2005.01822.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Language models are unsupervised multitask learners",
"authors": [
{
"first": "Alec",
"middle": [],
"last": "Radford",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Rewon",
"middle": [],
"last": "Child",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Luan",
"suffix": ""
},
{
"first": "Dario",
"middle": [],
"last": "Amodei",
"suffix": ""
},
{
"first": "Ilya",
"middle": [],
"last": "Sutskever",
"suffix": ""
}
],
"year": 2019,
"venue": "OpenAI Blog",
"volume": "1",
"issue": "8",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners. OpenAI Blog, 1(8):9.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Truth of varying shades: Analyzing language in fake news and political fact-checking",
"authors": [
{
"first": "Eunsol",
"middle": [],
"last": "Hannah Rashkin",
"suffix": ""
},
{
"first": "Jin",
"middle": [
"Yea"
],
"last": "Choi",
"suffix": ""
},
{
"first": "Svitlana",
"middle": [],
"last": "Jang",
"suffix": ""
},
{
"first": "Yejin",
"middle": [],
"last": "Volkova",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Choi",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "2931--2937",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hannah Rashkin, Eunsol Choi, Jin Yea Jang, Svit- lana Volkova, and Yejin Choi. 2017. Truth of vary- ing shades: Analyzing language in fake news and political fact-checking. In Proceedings of Empiri- cal Methods in Natural Language Processing, pages 2931-2937.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "Linguistic models for analyzing and detecting biased language",
"authors": [
{
"first": "Marta",
"middle": [],
"last": "Recasens",
"suffix": ""
},
{
"first": "Cristian",
"middle": [],
"last": "Danescu-Niculescu-Mizil",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Jurafsky",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1650--1659",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marta Recasens, Cristian Danescu-Niculescu-Mizil, and Dan Jurafsky. 2013. Linguistic models for an- alyzing and detecting biased language. In Proceed- ings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pa- pers), pages 1650-1659, Sofia, Bulgaria. Associa- tion for Computational Linguistics.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "Learning extraction patterns for subjective expressions",
"authors": [
{
"first": "Ellen",
"middle": [],
"last": "Riloff",
"suffix": ""
},
{
"first": "Janyce",
"middle": [],
"last": "Wiebe",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ellen Riloff and Janyce Wiebe. 2003. Learning extrac- tion patterns for subjective expressions. In Proceed- ings of Empirical Methods in Natural Language Pro- cessing.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "Fake news or truth? Using satirical cues to detect potentially misleading news",
"authors": [
{
"first": "Victoria",
"middle": [],
"last": "Rubin",
"suffix": ""
},
{
"first": "Niall",
"middle": [],
"last": "Conroy",
"suffix": ""
},
{
"first": "Yimin",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Sarah",
"middle": [],
"last": "Cornwell",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the second workshop on computational approaches to deception detection",
"volume": "",
"issue": "",
"pages": "7--17",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Victoria Rubin, Niall Conroy, Yimin Chen, and Sarah Cornwell. 2016. Fake news or truth? Using satirical cues to detect potentially misleading news. In Pro- ceedings of the second workshop on computational approaches to deception detection, pages 7-17.",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "The limitations of stylometry for detecting machine-generated fake news",
"authors": [
{
"first": "Tal",
"middle": [],
"last": "Schuster",
"suffix": ""
},
{
"first": "Roei",
"middle": [],
"last": "Schuster",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Darsh",
"suffix": ""
},
{
"first": "Regina",
"middle": [],
"last": "Shah",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Barzilay",
"suffix": ""
}
],
"year": 2020,
"venue": "Computational Linguistics",
"volume": "",
"issue": "",
"pages": "1--12",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tal Schuster, Roei Schuster, Darsh J Shah, and Regina Barzilay. 2020. The limitations of stylometry for detecting machine-generated fake news. Computa- tional Linguistics, pages 1-12.",
"links": null
},
"BIBREF31": {
"ref_id": "b31",
"title": "Automated readability index",
"authors": [
{
"first": "R",
"middle": [
"J"
],
"last": "Senter",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Edgar",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Smith",
"suffix": ""
}
],
"year": 1967,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "RJ Senter and Edgar A Smith. 1967. Automated read- ability index. Technical report.",
"links": null
},
"BIBREF32": {
"ref_id": "b32",
"title": "Joint learning of a dual SMT system for paraphrase generation",
"authors": [
{
"first": "Hong",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "Ming",
"middle": [],
"last": "Zhou",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "38--42",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hong Sun and Ming Zhou. 2012. Joint learning of a dual SMT system for paraphrase generation. In Pro- ceedings of the 50th Annual Meeting of the Associa- tion for Computational Linguistics (Volume 2: Short Papers), pages 38-42, Jeju Island, Korea. Associa- tion for Computational Linguistics.",
"links": null
},
"BIBREF33": {
"ref_id": "b33",
"title": "Sequence to sequence learning with neural networks",
"authors": [
{
"first": "Ilya",
"middle": [],
"last": "Sutskever",
"suffix": ""
},
{
"first": "Oriol",
"middle": [],
"last": "Vinyals",
"suffix": ""
},
{
"first": "Quoc V",
"middle": [],
"last": "Le",
"suffix": ""
}
],
"year": 2014,
"venue": "Advances in neural information processing systems",
"volume": "",
"issue": "",
"pages": "3104--3112",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ilya Sutskever, Oriol Vinyals, and Quoc V Le. 2014. Sequence to sequence learning with neural networks. In Advances in neural information processing sys- tems, pages 3104-3112.",
"links": null
},
"BIBREF34": {
"ref_id": "b34",
"title": "Attention is all you need",
"authors": [
{
"first": "Ashish",
"middle": [],
"last": "Vaswani",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Shazeer",
"suffix": ""
},
{
"first": "Niki",
"middle": [],
"last": "Parmar",
"suffix": ""
},
{
"first": "Jakob",
"middle": [],
"last": "Uszkoreit",
"suffix": ""
},
{
"first": "Llion",
"middle": [],
"last": "Jones",
"suffix": ""
},
{
"first": "Aidan",
"middle": [
"N"
],
"last": "Gomez",
"suffix": ""
},
{
"first": "\u0141ukasz",
"middle": [],
"last": "Kaiser",
"suffix": ""
},
{
"first": "Illia",
"middle": [],
"last": "Polosukhin",
"suffix": ""
}
],
"year": 2017,
"venue": "Advances in neural information processing systems",
"volume": "",
"issue": "",
"pages": "5998--6008",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, \u0141ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information pro- cessing systems, pages 5998-6008.",
"links": null
},
"BIBREF35": {
"ref_id": "b35",
"title": "Misleading or falsification: Inferring deceptive strategies and types in online news and social media",
"authors": [
{
"first": "Svitlana",
"middle": [],
"last": "Volkova",
"suffix": ""
},
{
"first": "Jin",
"middle": [
"Yea"
],
"last": "Jang",
"suffix": ""
}
],
"year": 2018,
"venue": "Companion Proceedings of the The Web Conference",
"volume": "",
"issue": "",
"pages": "575--583",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Svitlana Volkova and Jin Yea Jang. 2018. Misleading or falsification: Inferring deceptive strategies and types in online news and social media. In Compan- ion Proceedings of the The Web Conference 2018, pages 575-583.",
"links": null
},
"BIBREF36": {
"ref_id": "b36",
"title": "Separating facts from fiction: Linguistic models to classify suspicious and trusted news posts on Twitter",
"authors": [
{
"first": "Svitlana",
"middle": [],
"last": "Volkova",
"suffix": ""
},
{
"first": "Kyle",
"middle": [],
"last": "Shaffer",
"suffix": ""
},
{
"first": "Jin",
"middle": [
"Yea"
],
"last": "Jang",
"suffix": ""
},
{
"first": "Nathan",
"middle": [],
"last": "Hodas",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "647--653",
"other_ids": {
"DOI": [
"10.18653/v1/P17-2102"
]
},
"num": null,
"urls": [],
"raw_text": "Svitlana Volkova, Kyle Shaffer, Jin Yea Jang, and Nathan Hodas. 2017. Separating facts from fic- tion: Linguistic models to classify suspicious and trusted news posts on Twitter. In Proceedings of the 55th Annual Meeting of the Association for Compu- tational Linguistics (Volume 2: Short Papers), pages 647-653, Vancouver, Canada. Association for Com- putational Linguistics.",
"links": null
},
"BIBREF37": {
"ref_id": "b37",
"title": "The spread of true and false news online",
"authors": [
{
"first": "Soroush",
"middle": [],
"last": "Vosoughi",
"suffix": ""
},
{
"first": "Deb",
"middle": [],
"last": "Roy",
"suffix": ""
},
{
"first": "Sinan",
"middle": [],
"last": "Aral",
"suffix": ""
}
],
"year": 2018,
"venue": "Science",
"volume": "359",
"issue": "6380",
"pages": "1146--1151",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Soroush Vosoughi, Deb Roy, and Sinan Aral. 2018. The spread of true and false news online. Science, 359(6380):1146-1151.",
"links": null
},
"BIBREF38": {
"ref_id": "b38",
"title": "Liar, liar pants on fire\": A new benchmark dataset for fake news detection",
"authors": [
{
"first": "William",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Wang",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "422--426",
"other_ids": {
"DOI": [
"10.18653/v1/P17-2067"
]
},
"num": null,
"urls": [],
"raw_text": "William Yang Wang. 2017. \"Liar, liar pants on fire\": A new benchmark dataset for fake news detection. In Proceedings of the 55th Annual Meeting of the As- sociation for Computational Linguistics (Volume 2: Short Papers), pages 422-426, Vancouver, Canada. Association for Computational Linguistics.",
"links": null
},
"BIBREF39": {
"ref_id": "b39",
"title": "Information disorder: Toward an interdisciplinary framework for research and policy making",
"authors": [
{
"first": "Claire",
"middle": [],
"last": "Wardle",
"suffix": ""
},
{
"first": "Hossein",
"middle": [],
"last": "Derakhshan",
"suffix": ""
}
],
"year": 2017,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Claire Wardle and Hossein Derakhshan. 2017. Infor- mation disorder: Toward an interdisciplinary frame- work for research and policy making. Council of Europe report, 27.",
"links": null
},
"BIBREF40": {
"ref_id": "b40",
"title": "Google's neural machine translation system",
"authors": [
{
"first": "Yonghui",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Mike",
"middle": [],
"last": "Schuster",
"suffix": ""
},
{
"first": "Zhifeng",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Quoc",
"suffix": ""
},
{
"first": "Mohammad",
"middle": [],
"last": "Le",
"suffix": ""
},
{
"first": "Wolfgang",
"middle": [],
"last": "Norouzi",
"suffix": ""
},
{
"first": "Maxim",
"middle": [],
"last": "Macherey",
"suffix": ""
},
{
"first": "Yuan",
"middle": [],
"last": "Krikun",
"suffix": ""
},
{
"first": "Qin",
"middle": [],
"last": "Cao",
"suffix": ""
},
{
"first": "Klaus",
"middle": [],
"last": "Gao",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Macherey",
"suffix": ""
}
],
"year": 2016,
"venue": "Bridging the gap between human and machine translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1609.08144"
]
},
"num": null,
"urls": [],
"raw_text": "Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, et al. 2016. Google's neural machine translation system: Bridging the gap between hu- man and machine translation. arXiv preprint arXiv:1609.08144.",
"links": null
},
"BIBREF41": {
"ref_id": "b41",
"title": "Optimizing statistical machine translation for text simplification",
"authors": [
{
"first": "Wei",
"middle": [],
"last": "Xu",
"suffix": ""
},
{
"first": "Courtney",
"middle": [],
"last": "Napoles",
"suffix": ""
},
{
"first": "Ellie",
"middle": [],
"last": "Pavlick",
"suffix": ""
},
{
"first": "Quanze",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
}
],
"year": 2016,
"venue": "Transactions of the Association for Computational Linguistics",
"volume": "4",
"issue": "",
"pages": "401--415",
"other_ids": {
"DOI": [
"10.1162/tacl_a_00107"
]
},
"num": null,
"urls": [],
"raw_text": "Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch. 2016. Optimizing statistical machine translation for text simplification. Transactions of the Association for Computational Linguistics, 4:401-415.",
"links": null
},
"BIBREF42": {
"ref_id": "b42",
"title": "Automatic detection of deception in child-produced speech using syntactic complexity features",
"authors": [
{
"first": "Maria",
"middle": [],
"last": "Yancheva",
"suffix": ""
},
{
"first": "Frank",
"middle": [],
"last": "Rudzicz",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "944--953",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Maria Yancheva and Frank Rudzicz. 2013. Automatic detection of deception in child-produced speech us- ing syntactic complexity features. In Proceedings of the 51st Annual Meeting of the Association for Com- putational Linguistics (Volume 1: Long Papers), vol- ume 1, pages 944-953.",
"links": null
},
"BIBREF43": {
"ref_id": "b43",
"title": "Defending against neural fake news",
"authors": [
{
"first": "Rowan",
"middle": [],
"last": "Zellers",
"suffix": ""
},
{
"first": "Ari",
"middle": [],
"last": "Holtzman",
"suffix": ""
},
{
"first": "Hannah",
"middle": [],
"last": "Rashkin",
"suffix": ""
},
{
"first": "Yonatan",
"middle": [],
"last": "Bisk",
"suffix": ""
},
{
"first": "Ali",
"middle": [],
"last": "Farhadi",
"suffix": ""
},
{
"first": "Franziska",
"middle": [],
"last": "Roesner",
"suffix": ""
},
{
"first": "Yejin",
"middle": [],
"last": "Choi",
"suffix": ""
}
],
"year": 2019,
"venue": "Advances in Neural Information Processing Systems",
"volume": "",
"issue": "",
"pages": "9054--9065",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rowan Zellers, Ari Holtzman, Hannah Rashkin, Yonatan Bisk, Ali Farhadi, Franziska Roesner, and Yejin Choi. 2019. Defending against neural fake news. In Advances in Neural Information Process- ing Systems, pages 9054-9065.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Conceptual framework for understanding and explicitly measuring linguistic and stylistic properties of deception via generation and translation.",
"num": null,
"uris": null,
"type_str": "figure"
},
"FIGREF1": {
"text": "Conspiracy sources tend to explain an event or practice by the coordinated actions of powerful people.\u2022 Propaganda sources tend to persuade and manipulate public opinions and attitudes.\u2022 Hoax sources seek to fool the gullible. \u2022 Clickbait sources use attention-grabbing or vague headlines to drive engagement. \u2022 Credible sources provide factual information",
"num": null,
"uris": null,
"type_str": "figure"
},
"TABREF0": {
"text": "Conspiracy 1) radioactive material stolen in dallas raises concern over # radioactive weapons 2) iran [OOV] $ [D] billion on finding a giant submarine [OOV] Disinfo 1) # lavrov : we hope the # [OOV] is a long -term solution for russia ' s foreign -[OOV] partnership 2) blast in # ankara after explosions heard in turkey : reports Hoax 1) [D] times you ' re all confident you don ' t know what to see on your genes 2) girl loses her life in the world . . .",
"type_str": "table",
"content": "<table><tr><td/><td>1) # erdogan : so much # isis will be forced to enter # syria [URL] # syriacrisis</td></tr><tr><td>Propaganda</td><td>2) # putin : russia ' s # gas reserves soar at [OOV] since june [D][D][D][D]</td></tr></table>",
"num": null,
"html": null
},
"TABREF3": {
"text": "Encoder-decoder model performance for translations from deceptive to credible content and vice versa. MA refers to models with multiplicative attention and SA refers to models with self-attention.",
"type_str": "table",
"content": "<table><tr><td>DECEPTIVE SOURCE</td><td>CREDIBLE TARGET</td><td>MODEL OUTPUT</td><td>SARI</td></tr><tr><td>russian experts to fly to turkey to</td><td>russian ambassador to turkey at-</td><td>russian ambassador to turkey shot</td><td>0.64</td></tr><tr><td>investigate ambassadors murder</td><td>tacked at photo exhibition</td><td>dead at photo exhibition in ankara</td><td/></tr><tr><td>north korea claims to have suc-</td><td>north korea is believed to have con-</td><td>north korea conducts fifth and</td><td>0.41</td></tr><tr><td>cessfully carried out its fifth nu-</td><td>ducted a fifth nuclear test</td><td>largest nuclear test</td><td/></tr><tr><td>clear test</td><td/><td/><td/></tr></table>",
"num": null,
"html": null
},
"TABREF5": {
"text": "Linguistic differences between parallel content from credible and deceptive sources controlling for topic bias. Stat. sign. differences are shown in bold (Mann-Whitney U test p < 0.005).",
"type_str": "table",
"content": "<table/>",
"num": null,
"html": null
}
}
}
} |