File size: 83,109 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 |
{
"paper_id": "D07-1035",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T16:20:10.450407Z"
},
"title": "Low-Quality Product Review Detection in Opinion Summarization",
"authors": [
{
"first": "Jingjing",
"middle": [],
"last": "Liu",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Nankai University Tianjin",
"location": {
"country": "China"
}
},
"email": ""
},
{
"first": "Yunbo",
"middle": [],
"last": "Cao",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Microsoft Research Asia",
"location": {
"settlement": "Beijing",
"country": "China"
}
},
"email": "yucao@microsoft.com"
},
{
"first": "Chin-Yew",
"middle": [],
"last": "Lin",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Microsoft Research Asia",
"location": {
"settlement": "Beijing",
"country": "China"
}
},
"email": ""
},
{
"first": "Yalou",
"middle": [],
"last": "Huang",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Nankai University Tianjin",
"location": {
"country": "China"
}
},
"email": "huangyl@nankai.edu.cn"
},
{
"first": "Ming",
"middle": [],
"last": "Zhou",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Microsoft Research Asia",
"location": {
"settlement": "Beijing",
"country": "China"
}
},
"email": "mingzhou@microsoft.com"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Product reviews posted at online shopping sites vary greatly in quality. This paper addresses the problem of detecting lowquality product reviews. Three types of biases in the existing evaluation standard of product reviews are discovered. To assess the quality of product reviews, a set of specifications for judging the quality of reviews is first defined. A classificationbased approach is proposed to detect the low-quality reviews. We apply the proposed approach to enhance opinion summarization in a two-stage framework. Experimental results show that the proposed approach effectively (1) discriminates lowquality reviews from high-quality ones and (2) enhances the task of opinion summarization by detecting and filtering lowquality reviews.",
"pdf_parse": {
"paper_id": "D07-1035",
"_pdf_hash": "",
"abstract": [
{
"text": "Product reviews posted at online shopping sites vary greatly in quality. This paper addresses the problem of detecting lowquality product reviews. Three types of biases in the existing evaluation standard of product reviews are discovered. To assess the quality of product reviews, a set of specifications for judging the quality of reviews is first defined. A classificationbased approach is proposed to detect the low-quality reviews. We apply the proposed approach to enhance opinion summarization in a two-stage framework. Experimental results show that the proposed approach effectively (1) discriminates lowquality reviews from high-quality ones and (2) enhances the task of opinion summarization by detecting and filtering lowquality reviews.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In the past few years, there has been an increasing interest in mining opinions from product reviews (Pang, et al, 2002; Liu, et al, 2004; Popescu and Etzioni, 2005) . However, due to the lack of editorial and quality control, reviews on products vary greatly in quality. Thus, it is crucial to have a mechanism capable of assessing the quality of reviews and detecting low-quality/noisy reviews.",
"cite_spans": [
{
"start": 101,
"end": 120,
"text": "(Pang, et al, 2002;",
"ref_id": "BIBREF16"
},
{
"start": 121,
"end": 138,
"text": "Liu, et al, 2004;",
"ref_id": null
},
{
"start": 139,
"end": 165,
"text": "Popescu and Etzioni, 2005)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Some shopping sites already provide a function of assessing the quality of reviews. For example, Amazon 1 allows users to vote for the helpfulness of each review and then ranks the reviews based on the accumulated votes. However, according to our survey in Section 3, users' votes at Amazon have three kinds of biases as follows: (1) imbalance vote bias, (2) winner circle bias, and (3) early bird bias. Existing studies (Kim et al, 2006; Zhang and Varadarajan, 2006 ) used these users' votes for training ranking models to assess the quality of reviews, which therefore are subject to these biases.",
"cite_spans": [
{
"start": 421,
"end": 438,
"text": "(Kim et al, 2006;",
"ref_id": "BIBREF11"
},
{
"start": 439,
"end": 466,
"text": "Zhang and Varadarajan, 2006",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we demonstrate the aforementioned biases and define a standard specification to measure the quality of product reviews. We then manually annotate a set of ground-truth with real world product review data conforming to the specification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "To automatically detect low-quality product reviews, we propose a classification-based approach learned from the annotated ground-truth. The proposed approach explores three aspects of product reviews, namely informativeness, readability, and subjectiveness.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We apply the proposed approach to opinion summarization, a typical opinion mining task. The proposed approach enhances the existing work in a two-stage framework, where the low-quality review detection is applied right before the summarization stage.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Experimental results show that the proposed approach can discriminate low-quality reviews from high-quality ones effectively. In addition, the task of opinion summarization can be enhanced by detecting and filtering low-quality reviews.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The rest of the paper is organized as follows: Section 2 introduces the related work. In Section 3, we define the quality of product reviews. In Section 4, we present our approach to detecting lowquality reviews. In Section 5, we empirically verify the effectiveness of the proposed approach and its use for opinion summarization. Section 6 summarizes our work in this paper and points out the future work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The problem of evaluating helpfulness of reviews (Kim et al, 2006) , also known as learning utility of reviews (Zhang and Varadarajan, 2006) , is quite similar to our problem of assessing the quality of reviews.",
"cite_spans": [
{
"start": 49,
"end": 66,
"text": "(Kim et al, 2006)",
"ref_id": "BIBREF11"
},
{
"start": 111,
"end": 140,
"text": "(Zhang and Varadarajan, 2006)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluating Helpfulness of Reviews",
"sec_num": "2.1"
},
{
"text": "In practice, researchers in this area considered the problem as a ranking problem and solved it with regression models. In the process of model training and testing, they used the ground-truth derived from users' votes of helpfulness provided by Amazon. As we will show later in Section 3, these models all suffered from three types of voting bias.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluating Helpfulness of Reviews",
"sec_num": "2.1"
},
{
"text": "In our work, we avoid using users' votes by developing a specification on the quality of reviews and building a ground-truth according to the specification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluating Helpfulness of Reviews",
"sec_num": "2.1"
},
{
"text": "One area of research on opinion mining from product reviews is to judge whether a review expresses a positive or a negative opinion. For example, Turney (2006) presented a simple unsupervised learning algorithm in judging reviews as \"thumbs up\" (recommended) or \"thumbs down\" (not recommended). Pang et al (2002) considered the same problem and presented a set of supervised machine learning approaches to it. For other work see also Dave et al. (2003) , Lee (2004, 2005) .",
"cite_spans": [
{
"start": 295,
"end": 312,
"text": "Pang et al (2002)",
"ref_id": "BIBREF16"
},
{
"start": 434,
"end": 452,
"text": "Dave et al. (2003)",
"ref_id": "BIBREF1"
},
{
"start": 455,
"end": 471,
"text": "Lee (2004, 2005)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Mining Opinions from Reviews",
"sec_num": "2.2"
},
{
"text": "Another area of research on opinion mining is to extract and summarize users' opinions from product reviews (Hu and Liu, 2004; Liu et al., 2005; Popescu and Etzioni, 2005) . Typically, a sentence or a text segment in the reviews is treated as the basic unit. The polarity of users' sentiments on a product feature in each unit is extracted. Then the aggregation of the polarities of individual senti-ments is presented to users so that they can have an at-a-glance view on how other experienced users rated on a certain product. The major weakness in the existing studies is that all the reviews, including low-quality ones, are taken into consideration and treated equally for generating the summary. In this paper, we enhance the application by detecting and filtering low-quality reviews. In order to achieve that, we first define what the quality of reviews is.",
"cite_spans": [
{
"start": 108,
"end": 126,
"text": "(Hu and Liu, 2004;",
"ref_id": null
},
{
"start": 127,
"end": 144,
"text": "Liu et al., 2005;",
"ref_id": "BIBREF13"
},
{
"start": 145,
"end": 171,
"text": "Popescu and Etzioni, 2005)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Mining Opinions from Reviews",
"sec_num": "2.2"
},
{
"text": "In this section, we will first show three biases of users' votes observed on Amazon, and then present our specification on the quality of product reviews.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Quality of Product Reviews",
"sec_num": "3"
},
{
"text": "In our study, we use the product reviews on digital cameras crawled from Amazon as our data set. The data set consists of 23,141 reviews on 946 digital cameras. At the Amazon site, users could vote for a review with a \"helpful\" or \"unhelpful\" label. Thus, for each review there are two numbers indicating the statistics of these two labels, namely the number of \"helpful\" votes and that of \"unhelpful\" ones. Kim et al (2006) used the percentage of \"helpful\" votes as the measure of evaluating the \"quality of reviews\" in their experiments. We call the ground-truth based on this measure as \"Amazon ground-truth\".",
"cite_spans": [
{
"start": 408,
"end": 424,
"text": "Kim et al (2006)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Amazon Ground-truth",
"sec_num": "3.1"
},
{
"text": "Certainly, the ground-truth has the advantage of convenience. However, we identify three types of biases that make the Amazon ground-truth not always suitable for determining the quality of reviews. We describe these biases in details in the rest of this section. At the Amazon site, users tend to value others' opinions positively rather than negatively. From Figure 1 , we can see that a half of the 23,141 reviews (corresponding to the two bars on the right of the figure) have more than 90% \"helpful\" votes, including 9,100 reviews with 100% \"helpful\" votes. From an in-depth investigation on these highly-voted reviews, we observed that some did not really have as good quality as the votes hint. For example, in Figure 2 , the review about Canon PowerShot S500 receives 40 \"helpful\" votes out of 40 votes although it only gives very brief description on the product features in its second paragraph. We call this type of bias \"imbalance vote\" bias. Figure 3 . Votes of the top-50 ranked reviews There also exists a bootstrapping effect of \"hot\" reviews at the Amazon site. Figure 3 shows the \"helpful\" votes for the top 50 ranked reviews. The numbers are averaged over 127 digital cameras which have no less than 50 reviews. As shown in this figure, the top two reviews hold more than 250 and 140 votes respectively on average; while the numbers of votes held by lower-ranked reviews decrease exponentially. This is so-called the \"winner circle\" bias: the more votes a review gains, the more default authority it would appear to the readers, which in turn will influence the objectivity of the readers' votes. Also, the higher ranked reviews would attract more eyeballs and therefore gain more people's votes. This mutual influence among labelers should be avoided when the votes are used as the evaluation standard.",
"cite_spans": [],
"ref_spans": [
{
"start": 361,
"end": 369,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 718,
"end": 726,
"text": "Figure 2",
"ref_id": "FIGREF2"
},
{
"start": 955,
"end": 963,
"text": "Figure 3",
"ref_id": null
},
{
"start": 1079,
"end": 1087,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Amazon Ground-truth",
"sec_num": "3.1"
},
{
"text": "Publication date can influence the accumulation of users' votes. In Figure 4 , the n'th publication date represents the n'th month after the product is released. The number in the figure is averaged over all the digital cameras in the data set. We can observe a clear trend that the earlier a review is posted, the more votes it will get. This is simply because reviews posted earlier are exposed to users for a longer time. Therefore, some high quality reviews may get fewer users' vote because of later publication. We call this \"early bird\" bias.",
"cite_spans": [],
"ref_spans": [
{
"start": 68,
"end": 76,
"text": "Figure 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Figure 4. Dependency on publication date",
"sec_num": null
},
{
"text": "Besides these aforementioned biases, using the raw rating from readers directly also fails to provide a clear guideline for what a good review consists of. In this section, we provide such a guideline, which we name as the specification (SPEC).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Specification of Quality",
"sec_num": "3.2"
},
{
"text": "In the SPEC, we define four categories of review quality which represent different values of the reviews to users' purchase decision: \"best review\", \"good review\", \"fair review\", and \"bad review\". A generic description of the SPEC is as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Specification of Quality",
"sec_num": "3.2"
},
{
"text": "A best review must be a rather complete and detailed comment on a product. It presents several aspects of a product and provides convincing opinions with enough evidence. Usually a best review could be taken as the main reference that users only need to read before making their purchase decision on a certain product. The first review in Figure 5 is a best review. It presents several product features and provides convincing opinions with sufficient evidence. It is also in a good format for readers to easily understand. Note that we omit some words in the example to save the space. A good review is a relatively complete comment on a product, but not with as much supporting evidence as necessary. It could be used as a strong and influential reference, but not as the only recommendation. The second review in Figure 5 is such an example.",
"cite_spans": [
{
"start": 339,
"end": 345,
"text": "Figure",
"ref_id": null
}
],
"ref_spans": [
{
"start": 816,
"end": 824,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Specification of Quality",
"sec_num": "3.2"
},
{
"text": "A fair review contains a very brief description on a product. It does not supply detailed evaluation on the product, but only comments on some aspects of the product. For example, the third review in Figure 5 mainly talks about \"the delay between pictures\", but less about other aspects of the camera.",
"cite_spans": [],
"ref_spans": [
{
"start": 200,
"end": 208,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Specification of Quality",
"sec_num": "3.2"
},
{
"text": "A bad review is usually an incorrect description of a product with misleading information. It talks little about a specific product but much about some general topics (e.g. photography). For example, the last review in Figure 5 talks about the topic of \"generic battery\", but does not specify any digital camera. A bad review is an \"unhelpful\" review that can be ignored. ",
"cite_spans": [],
"ref_spans": [
{
"start": 219,
"end": 227,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Specification of Quality",
"sec_num": "3.2"
},
{
"text": "According to the SPEC defined above, we built a ground-truth from the Amazon data set. We randomly selected 100 digital cameras and 50 reviews for each camera. Totally we have 4,909 reviews since some digital cameras have fewer than 50 unique reviews. Then we hired two annotators to label the reviews with the SPEC as their guideline. As the result, we have two independent copies of annotations on 4,909 reviews, with the labels of \"best\", \"good\", \"fair\", and \"bad\". Table 1 shows the confusion matrix between the two copies of annotation. The value of the kappa statistic (Cohen, 1960) calculated from the matrix is 0.8142. This shows that the two annotators achieved highly consistent results by following the SPEC, although they worked independently. In order to examine the difference between our annotations and Amazon ground-truth, we evaluate the Amazon ground-truth against the annotations, with the measure of \"error rate of preference pairs\" (Herbrich et al, 1999) .",
"cite_spans": [
{
"start": 575,
"end": 588,
"text": "(Cohen, 1960)",
"ref_id": "BIBREF0"
},
{
"start": 954,
"end": 976,
"text": "(Herbrich et al, 1999)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [
{
"start": 469,
"end": 476,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Annotation of Quality",
"sec_num": "3.3"
},
{
"text": "= | | | | (1)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation of Quality",
"sec_num": "3.3"
},
{
"text": "where the \"preference pair\" is defined as a pair of reviews with a order. For example, a best review and a good review correspond to a preference pair with the order of \"best review preferring to good review\". The \"all preference pairs\" are collected from one of the annotations (the annotation 1 or the annotation 2) by ignoring the pairs from the same category. The \"incorrect preference pairs\" are the preference pairs collected from the Amazon ground-truth but not with the same order as that in the all preference pairs. The order of the preference pair collected from the Amazon ground-truth is evaluated on the basis of the percentage score as described in Section 3.1. The error rate of preference pairs based on the annotation 1 and that based on the annotation 2 are 0.448 and 0.446, respectively, averaged over 100 digital cameras. The high error rate of preference pairs demonstrates that the Amazon ground-truth diverges from the annotations (our ground-truth) significantly.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation of Quality",
"sec_num": "3.3"
},
{
"text": "To discover which kind of ground-truth is more reasonable, we ask an additional annotator (the third annotator) to compare these two kinds of ground-truth. More specifically, we randomly selected 100 preference pairs whose orders the two kinds of ground-truth don't agree on (called incorrect preference pairs in the evaluation above). As for our ground-truth, we choose the Annotation 1 in the new test. Then, the third annotator is asked to assign a preference order for each selected pair. Note that the third annotator is blind to both our specification and the existing preference order. Last, we evaluate the two kinds of ground-truth with the new annotation. Among 100 pairs, our ground-truth agrees to the new annotation on 85 pairs while the Amazon ground-truth agrees to the new annotation on 15 pairs. To confirm the result, yet another annotator (the fourth annotator) is called to repeat the same annotation independently as the third one. And we obtain the same statistical result (85 vs. 15) although the fourth annotator does not agree with the third annotator on some pairs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation of Quality",
"sec_num": "3.3"
},
{
"text": "In practice, we treat the reviews in the first three categories (\"best\", \"good\" and \"fair\") as highquality reviews and those in the \"bad\" category as low-quality reviews, since our goal is to identify low quality reviews that should not be considered when creating product review summaries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation of Quality",
"sec_num": "3.3"
},
{
"text": "We employ a statistical machine learning approach to address the problem of detecting low-quality products reviews.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Classification of Product Reviews",
"sec_num": "4"
},
{
"text": "Given a training data set = , 1 , we construct a model that can minimize the error in prediction of y given x (generalization error). Here \u2208 and = { , } represents a product review and a label, respectively. When applied to a new instance x, the model predicts the corresponding y and outputs the score of the prediction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Classification of Product Reviews",
"sec_num": "4"
},
{
"text": "In our study, we focus on differentiating lowquality product reviews from high-quality ones. Thus, we treat the task as a binary classification problem.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Learning Model",
"sec_num": "4.1"
},
{
"text": "We employ SVM (Support Vector Machines) (Vapnik, 1995) as the model of classification. Given an instance x (product review), SVM assigns a score to it based on = +",
"cite_spans": [
{
"start": 40,
"end": 54,
"text": "(Vapnik, 1995)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Learning Model",
"sec_num": "4.1"
},
{
"text": "(2) where w denotes a vector of weights and b denotes an intercept. The higher the value of f(x) is, the higher the quality of the instance x is. In classification, the sign of f(x) is used. If it is positive, then x is classified into the positive category (high-quality reviews), otherwise into the negative category (low-quality reviews).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Learning Model",
"sec_num": "4.1"
},
{
"text": "The construction of SVM needs labeled training data (in our case, the categories are \"high-quality reviews\" and \"low-quality reviews\"). Briefly, the learning algorithm creates the \"hyper plane\" in (2), such that the hyper plane separates the positive and negative instances in the training data with the largest \"margin\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Learning Model",
"sec_num": "4.1"
},
{
"text": "Product features (e.g., \"image quality\" for digital camera) in a review are good indicators of review quality. However, different product features may refer to the same meaning (e.g., \"battery life\" and \"power\"), which will bring redundancy in the study. In this paper, we formulize the problem as the \"resolution of product features\". Thus, the problem is reduced to how to determine the equivalence of a product feature in different forms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Product Feature Resolution",
"sec_num": "4.2"
},
{
"text": "In (Hu and Liu, 2004) , the matching of different product features is mentioned briefly and addressed by fuzzy matching. However, there exist many cases where the method fails to match the multiple mentions, e.g., \"battery life\" and \"power\", because it only considers string similarity. In this paper we propose to resolve the problem by leveraging two kinds of evidence: one is \"surface string\" evidence, the other is \"contextual evidence\".",
"cite_spans": [
{
"start": 3,
"end": 21,
"text": "(Hu and Liu, 2004)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Product Feature Resolution",
"sec_num": "4.2"
},
{
"text": "We use edit distance (Ukkonen, 1985) to compare the similarity between the surface strings of two mentions, and use contextual similarity to reflect the semantic similarity between two mentions.",
"cite_spans": [
{
"start": 21,
"end": 36,
"text": "(Ukkonen, 1985)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Product Feature Resolution",
"sec_num": "4.2"
},
{
"text": "When using contextual similarity, we split all the reviews into sentences. For each mention of a product feature, we take it as a query and search for all the relevant sentences. Then we construct a vector for the mention, by taking each unique term in the relevant sentences as a dimension of the vector. The cosine similarity between two vectors of mentions is then present to measure the contextual similarity between two mentions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Product Feature Resolution",
"sec_num": "4.2"
},
{
"text": "To detect low-quality reviews, our proposed approach explores three aspects of product reviews, namely informativeness, subjectiveness, and readability. We denote the features employed for learning as \"learning features\", discriminative from the \"product features\" we discussed above.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Feature Development for Learning",
"sec_num": "4.3"
},
{
"text": "As for informativeness, the resolution of product features is employed when we generate the learning features as listed below. Pairs mapping to the same product feature will be treated as the same product feature, when we calculate the frequency and the number of product features. We apply the approach proposed in (Hu and Liu, 2004) to extract product features.",
"cite_spans": [
{
"start": 316,
"end": 334,
"text": "(Hu and Liu, 2004)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Features on Informativeness",
"sec_num": "4.3.1"
},
{
"text": "We also use a list of product names and a list of brand names to generate the learning features. Both lists can be collected from the Amazon site because they are relatively stable within a time interval.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Features on Informativeness",
"sec_num": "4.3.1"
},
{
"text": "The learning features on the informativeness of a review are as follows. \uf0d8 Sentence level (SL) \uf0b7 The number of sentences in the review \uf0b7 The average length of sentences \uf0b7 The number of sentences with product features \uf0d8 Word level (WL) \uf0b7 The number of words in the review \uf0b7 The number of products (e.g., DMC-FZ50, EX-Z1000) in the review \uf0b7 The number of products in the title of a review \uf0b7 The number of brand names (e.g., Canon, Sony) in the review \uf0b7 The number of brand names in the title of a review \uf0d8 Product feature level (PFL) \uf0b7 The number of product features in the review \uf0b7 The total frequency of product features in the review \uf0b7 The average frequency of product features in the review \uf0b7 The number of product features in the title of a review \uf0b7 The total frequency of product features in the title of a review",
"cite_spans": [
{
"start": 422,
"end": 434,
"text": "Canon, Sony)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Features on Informativeness",
"sec_num": "4.3.1"
},
{
"text": "We make use of several features at paragraph level which indicate the underlying structure of the reviews. These features include, \uf0b7 The number of paragraphs in the review \uf0b7 The average length of paragraphs in the review \uf0b7 The number of paragraph separators in the review Here, we refer to the keywords, such as \"Pros\" vs. \"Cons\" as \"paragraph separators\". The keywords usually appear at the beginning of paragraphs for categorizing two contrasting aspects of a product. We extract the nouns and noun phrases at the beginning of each paragraph from the 4,909 reviews and use the most frequent 30 pairs of keywords as paragraph separators. Table 2 provides some examples of the extracted separators. ",
"cite_spans": [],
"ref_spans": [
{
"start": 639,
"end": 646,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Features on Readability",
"sec_num": "4.3.2"
},
{
"text": "We also take the subjectiveness of reviews into consideration. Unlike previous work (Kim et al, 2006; Zhang and Varadarajan, 2006) using shallow syntactic information directly, we use a sentiment analysis tool (Hu and Liu, 2004) which aggregates a set of shallow syntactic information. The tool is a classifier capable of determining the sentiment polarity of each sentence. We create three learning features regarding the subjectiveness of reviews. \uf0b7 The percentage of positive sentences in the review \uf0b7 The percentage of negative sentences in the review \uf0b7 The percentage of subjective sentences (regardless of positive or negative) in the review",
"cite_spans": [
{
"start": 84,
"end": 101,
"text": "(Kim et al, 2006;",
"ref_id": "BIBREF11"
},
{
"start": 102,
"end": 130,
"text": "Zhang and Varadarajan, 2006)",
"ref_id": "BIBREF21"
},
{
"start": 210,
"end": 228,
"text": "(Hu and Liu, 2004)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Features on Subjectiveness",
"sec_num": "4.3.3"
},
{
"text": "In this section, we describe our experiments with the proposed classification-based approach to lowquality review detection, and its effectiveness on the task of opinion summarization.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "5"
},
{
"text": "In our proposed approach, the problem of assessing quality of reviews is formalized as a binary classification problem. We conduct experiments by taking reviews in the categories of \"best\", \"good\", and \"fair\" as high-quality reviews and those in the \"bad\" category as low-quality reviews. As for classification model, we utilize the SVMLight toolkit (Joachims, 2004) . We randomly divide the 100 queries of digital cameras into two sets, namely a training set of 50 queries and a test set of 50 queries. For the two copies of annotations, we use the same division. We use the training set from \"annotation 1\" to train the model and apply the model to the test sets from both \"annotation 1\" and \"annotation 2\", respectively. Table 3 reports the accuracies of our approach to review classification. The accuracy is defined as the percentage of correctly classified reviews.",
"cite_spans": [
{
"start": 350,
"end": 366,
"text": "(Joachims, 2004)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [
{
"start": 724,
"end": 731,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Detecting Low-quality Reviews",
"sec_num": "5.1"
},
{
"text": "We take the approach that utilizes only the category of features on sentence level (SL) as the baseline, and incrementally add other categories of features on informativeness, readability and subjectiveness. We can see that both the features on word level (WL) and those on product feature level (PFL) can improve the performance of classification much. The features on readability can still increase the accuracy although the contribution is much less. The features on subjectiveness, however, make no contribution. Table 3 . Low-quality reviews detection We also conduct a more detailed analysis on each individual feature. Two categories of features on \"title\" and \"brand name\" have poor performance, which is due to the lack of information in the title and the low coverage of brand names in a review, respectively.",
"cite_spans": [],
"ref_spans": [
{
"start": 517,
"end": 524,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Detecting Low-quality Reviews",
"sec_num": "5.1"
},
{
"text": "One potential application of low-quality review detection is the opinion summarization of reviews.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "The process of opinion summarization of reviews with regards to a query of a product consists of the following steps (Liu et al, 2005) : 1. From each of the reviews, identify every text segment with opinion in the review, and determine the polarities of the opinion segments.",
"cite_spans": [
{
"start": 117,
"end": 134,
"text": "(Liu et al, 2005)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "2. For each product feature, generate a positive opinion set and a negative opinion set of opinion segments, denoted as POS( ) and NOS( ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "3. For each product feature, aggregate the numbers of segments in POS( ) andNOS( ), as opinion summarization on the product feature.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "In this process, all the reviews contribute the same. However, different reviews do hold different authorities. A positive/negative opinion from a high-quality review should not have the same weight as that from a low-quality review.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "We use a two-stage approach to enhance the reliability of summarization. That is, we add a process of low-quality review detection before the summarization process, so that the summarization result is obtained based on the high-quality reviews only. We are to demonstrate how much difference the proposed two-stage approach can bring into the opinion summarization.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "We use the best classification model trained as described in Section 5.1 to filter low-quality reviews, and do summarization on the high-quality reviews associated to the 50 test queries. We denote the proposed approach and the old approach as \"two-stage\" and \"one-stage\", respectively. Due to the limited space, we only give a visual comparison of the two approaches on \"image quality\" in Figure 6 . The upper figure shows the summarization of positive opinions and the lower figure shows that of negative opinions. From the figures we can see that the two-stage approach preserves fewer text segments as the result of filtering out many low-quality product reviews.",
"cite_spans": [],
"ref_spans": [
{
"start": 390,
"end": 398,
"text": "Figure 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "Summarizing Sentiments of Reviews",
"sec_num": "5.2"
},
{
"text": "To show the comparison on more features in a compressed space, we give the statistic ratio of change between two approaches instead. As for the evaluation measure, we define \"RatioOfChange\" (ROC) on a feature f as, ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Figure 6. Summarization on \"image quality\"",
"sec_num": null
},
{
"text": "where Rate * (f) is defined as, Table 4 shows some statistic results on ROC on five product features, namely \"image quality\"(IQ), \"battery\", \"LCD screen\" (LCD), \"flash\" and \"movie mode\" (MM). The values in the cells are the percentage of queries whose ROC is larger/smaller than the respective thresholds. We can see that a large portion of queries have big changes on the values of ROC. This means that the result achieved by the two-stage approach is substantially different from that achieved by the one-stage approach. There is no standard way to evaluate the quality of opinion summarization as it is rather a subjective problem. In order to demonstrate the impact of the two-stage approach, we turn to external authoritative sources other than Amazon.com as the objective evaluation reference. We observe that CNET 2 provides a professional \"editor's review\" for many products, which gives a rating in the range of 1~10 on product features. 9 digital cameras out of the 50 test queries are found to have the editor's rating on \"image quality\" at CNET. We use this rating to compare with the results of our opinion summarization. We rescale the Rate scores obtained by both the one-stage approach and the two-stage approach into the range of 1-10 in order to perform the comparison. Figure 7 provides the visual comparison. We can see that the result achieved by the two-stage approach has a much better (closer) resemblance to CNET rating than one-stage approach does. This indicates that our two-stage approach can achieve a more consistent summarization result to the professional evaluations by the editors. Although the CNET rating is not the absolute standard for product evaluation, it provides a professional yet objective evaluation of the products. Therefore, the experimental results demonstrate that our proposed approach could achieve more reliable opinion summarization which is closer to the generic evaluation from authoritative sources. ",
"cite_spans": [],
"ref_spans": [
{
"start": 32,
"end": 39,
"text": "Table 4",
"ref_id": null
},
{
"start": 1288,
"end": 1296,
"text": "Figure 7",
"ref_id": "FIGREF6"
}
],
"eq_spans": [],
"section": "Figure 6. Summarization on \"image quality\"",
"sec_num": null
},
{
"text": "Rate * ( ) = |POS( )| |POS( )| + |NOS( )| (4)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Figure 6. Summarization on \"image quality\"",
"sec_num": null
},
{
"text": "In this paper, we studied the problem of detecting low-quality product reviews. Our contribution can be summarized in two-fold: (1) we discovered three types of biases in the ground-truth used extensively in the existing work, and proposed a specification on the quality of product reviews. The three biases that we discovered are imbalance vote bias, winner circle bias, and early bird bias. 2Rooting on the new ground-truth (conforming to the proposed specification), we proposed a classification-based approach to low-quality product review detection, which yields better performance of opinion summarization. We hope to explore our future work in several areas, such as further consolidating the new ground-truth from different points of view and verifying the effectiveness of low-quality review detection with other applications. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "http://www.amazon.com",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 # Reviews Percentage of 'helpful' votes",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "A coefficient of agreement for nominal scales",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Cohen",
"suffix": ""
}
],
"year": 1960,
"venue": "Educational and Psychological Measurement",
"volume": "20",
"issue": "",
"pages": "37--46",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jacob Cohen. 1960. A coefficient of agreement for no- minal scales, Educational and Psychological Mea- surement 20: 37-46.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Mining the peanut gallery: opinion extraction and semantic classification of product reviews",
"authors": [
{
"first": "Kushal",
"middle": [],
"last": "Dave",
"suffix": ""
},
{
"first": "Steve",
"middle": [],
"last": "Lawrence",
"suffix": ""
},
{
"first": "David",
"middle": [
"M"
],
"last": "Pennock",
"suffix": ""
}
],
"year": 2003,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kushal Dave, Steve Lawrence, and David M. Pennock. 2003. Mining the peanut gallery: opinion extraction and semantic classification of product reviews. WWW'03.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Support vector regression machines",
"authors": [
{
"first": "Harris",
"middle": [],
"last": "Drucker",
"suffix": ""
},
{
"first": "Chris",
"middle": [
"J C"
],
"last": "Burges Linda Kaufman",
"suffix": ""
},
{
"first": "Alex",
"middle": [],
"last": "Smola",
"suffix": ""
},
{
"first": "Vladimir",
"middle": [],
"last": "Vapnik",
"suffix": ""
}
],
"year": 1997,
"venue": "Advances in Neural Information Processing Systems",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Harris Drucker, Chris J.C., Burges Linda Kaufman, Alex Smola and Vladimir Vapnik. 1997. Support vector regression machines. Advances in Neural In- formation Processing Systems.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "WordNet: an Electronic Lexical Database",
"authors": [
{
"first": "Christiane",
"middle": [],
"last": "Fellbaum",
"suffix": ""
}
],
"year": 1998,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christiane Fellbaum. 1998. WordNet: an Electronic Lexical Database, MIT Press.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Support Vector Learning for Ordinal Regression",
"authors": [
{
"first": "Ralf",
"middle": [],
"last": "Herbrich",
"suffix": ""
},
{
"first": "Thore",
"middle": [],
"last": "Graepel",
"suffix": ""
},
{
"first": "Klaus",
"middle": [],
"last": "Obermayer",
"suffix": ""
}
],
"year": 1999,
"venue": "Proc. of the 9th International Conference on Artificial Neural Networks",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ralf Herbrich, Thore Graepel, and Klaus Obermayer. 1999. Support Vector Learning for Ordinal Regres- sion. In Proc. of the 9th International Conference on Artificial Neural Networks.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Mining and Summarizing Customer Reviews",
"authors": [
{
"first": "Minqing",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "Bing",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Minqing Hu and Bing Liu. 2004a. Mining and Summa- rizing Customer Reviews. KDD'04.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Mining Opinion Features in Customer Reviews",
"authors": [
{
"first": "Minqing",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "Bing",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Minqing Hu and Bing Liu. 2004b. Mining Opinion Fea- tures in Customer Reviews. AAAI'04.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "IR: evaluation methods for retrieving highly relevant documents",
"authors": [],
"year": 2000,
"venue": "Kalervo Jarvelin & Jaana Kekalainen",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kalervo Jarvelin & Jaana Kekalainen. 2000. IR: evalua- tion methods for retrieving highly relevant docu- ments. SIGIR'00.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Identifying Comparative Sentences in Text Documents",
"authors": [
{
"first": "Nitin",
"middle": [],
"last": "Jindal",
"suffix": ""
},
{
"first": "Bing",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2006,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nitin Jindal and Bing Liu. 2006. Identifying Compara- tive Sentences in Text Documents. SIGIR'06.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Mining comparative sentences and relations",
"authors": [
{
"first": "Nitin",
"middle": [],
"last": "Jindal",
"suffix": ""
},
{
"first": "Bing",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2006,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nitin Jindal and Bing Liu. 2006. Mining comparative sentences and relations. AAAI'06.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "SVMlight --Support Vector Machine",
"authors": [
{
"first": "Thorsten",
"middle": [],
"last": "Joachims",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Thorsten Joachims. SVMlight --Support Vector Ma- chine. http://svmlight.joachims.org/, 2004.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Automatically Assessing Review Helpfulness",
"authors": [
{
"first": "Soo-Min",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "Patrick",
"middle": [],
"last": "Pantel",
"suffix": ""
},
{
"first": "Tim",
"middle": [],
"last": "Chklovski",
"suffix": ""
},
{
"first": "Marco",
"middle": [],
"last": "Pennacchiotti",
"suffix": ""
}
],
"year": 2006,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Soo-Min Kim, Patrick Pantel, Tim Chklovski, Marco Pennacchiotti. 2006. Automatically Assessing Re- view Helpfulness. EMNLP'06.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Automatic retrieval and clustering of similar words",
"authors": [
{
"first": "Dekang",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 1998,
"venue": "COLING-ACL'98",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dekang Lin. 1998, Automatic retrieval and clustering of similar words. COLING-ACL'98.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Opinion observer: analyzing and comparing opinions on the web. WWW '05",
"authors": [
{
"first": "Bing",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Minqing",
"middle": [],
"last": "Hu",
"suffix": ""
},
{
"first": "Junsheng",
"middle": [],
"last": "Cheng",
"suffix": ""
}
],
"year": 2005,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bing Liu, Minqing Hu, and Junsheng Cheng. 2005. Opinion observer: analyzing and comparing opinions on the web. WWW '05.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts",
"authors": [
{
"first": "Bo",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "Lillian",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bo Pang and Lillian Lee. 2004. A sentimental educa- tion: Sentiment analysis using subjectivity summari- zation based on minimum cuts. ACL'04.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales",
"authors": [
{
"first": "Bo",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "Lillian",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2005,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bo Pang and Lillian Lee. 2005. Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales. ACL'05.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Thumbs up? sentiment classification using machine learning techniques",
"authors": [
{
"first": "Bo",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "Lillian",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Vaithyanathan",
"suffix": ""
}
],
"year": 2002,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bo Pang and Lillian Lee, and S. Vaithyanathan. 2002. Thumbs up? sentiment classification using machine learning techniques. EMNLP'02.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Extracting product features and opinions from reviews",
"authors": [
{
"first": "Ana-Maria",
"middle": [],
"last": "Popescu",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Etzioni",
"suffix": ""
}
],
"year": 2005,
"venue": "HLT-EMNLP'05",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ana-Maria Popescu and O Etzioni. 2005. Extracting product features and opinions from reviews. HLT- EMNLP'05.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Thumbs up or thumbs down?: semantic orientation applied to unsupervised classification of reviews",
"authors": [
{
"first": "D",
"middle": [],
"last": "Peter",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Turney",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Peter D. Turney. 2001. Thumbs up or thumbs down?: semantic orientation applied to unsupervised classifi- cation of reviews. ACL'02",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Algorithms for approximate string matching. Information and Control",
"authors": [
{
"first": "",
"middle": [],
"last": "Esko Ukkonen",
"suffix": ""
}
],
"year": 1985,
"venue": "",
"volume": "",
"issue": "",
"pages": "100--118",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Esko Ukkonen. 1985. Algorithms for approximate string matching. Information and Control, pp. 100 -118.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "The Nature of Statistical Learning Theory",
"authors": [
{
"first": "Vladimir",
"middle": [
"N"
],
"last": "Vapnik",
"suffix": ""
}
],
"year": 1995,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Vladimir N. Vapnik. 1995. The Nature of Statistical Learning Theory. Springer.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Utility Scoring of Product Reviews",
"authors": [
{
"first": "Zhu",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Balaji",
"middle": [],
"last": "Varadarajan",
"suffix": ""
}
],
"year": 2006,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhu Zhang and Balaji Varadarajan. 2006. Utility Scor- ing of Product Reviews. CIKM'06",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Reviews' percentage scores"
},
"FIGREF2": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "An example review"
},
"FIGREF4": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Figure 5. Example reviews"
},
"FIGREF5": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "2 http://www.cnet.com"
},
"FIGREF6": {
"num": null,
"type_str": "figure",
"uris": null,
"text": "Comparison with CNET rating"
},
"TABREF1": {
"html": null,
"type_str": "table",
"text": "Fuji, and More. The new release of 5.0 mega pixels has shot prices for digital cameras up to $1000+. This camera I purchased through a Private Dealer cost me $400.86. The Retail Price is Running $499.00 to $599.00. Purchase this camera from a wholesale dealer for the best price $377.00. Great Photo Even in dim light w/o a flash. The p10 is very compact. Can easily fit into any pocket. The camera can record 90 minutes of mpeg like a home movie. There are a lot of great digital cameras on the market that shoot good pictures and video. What makes the p10 the top pick isit comes with a rechargeable lithium battery. Many use AA batteries, the digital camera consumes theses AA batteries in about two hours time while the unit is on. That can add continuous expense to the camera. It's also the best resolution on the market. 6.0 megapix is out, though only a few. And the smallest that we found. Also the best price for a major brand. There is nothing wrong with the 2100 except for the very noticeable delay between pics. The camera's digital processor takes about 5 seconds after a photo is snapped to ready itself for the next one. Otherwise, the optics, the 3X optical zoom and the 2 megapixel resolution are fine for anything from Internet apps to 8\" x 10\" print enlarging. It is competent, not spectacular, but it gets the job done at an agreeable price point.",
"num": null,
"content": "<table><tr><td>Best Review: I purchased this camera about six months ago after my Kodak Easyshare camera completely died on me. I did a little research and read only good things about this Canon camera so I decided to go with it because it was very reasonably priced (about $200). Not only did the camera live up to my expectations, it surpassed them by leaps and bounds! Here are the things I have loved about this camera: BATTERY -this camera has the best battery of any digital cam-era I have ever owned or used. \u2026 EASY TO USE -I was able to \u2026 PICTURE QUALITY -all of the pictures I've taken and printed out have been great. \u2026 FEATURES -I love the ability to quickly and easily \u2026 LCD SCREEN -I was hoping \u2026 SD MEMORY CARD -I was also looking for a camera that used SD memory cards. Mostly because\u2026 I cannot stress how highly I recommend this camera. I will never buy another digital camera besides Canon again. And the A610 (as well as the A620 -the 7.0MP version) is the best digital camera I've ever used. Good Review: The Sony DSC \"P10\" Digital Camera is the top pick for CSC. Running against cameras like Olympus stylus, Canon Powereshot, Sony V1, Nikon, Fair Review: Bad Review: I want to point out that you should never buy a generic battery, like the person from San Diego who reviewed the S410 on May 15, 2004, was recommending. Yes you'd save money, but there have been many reports of generic batteries exploding when charged for too long. And don't think if your generic battery explodes you can sue somebody and win millions. These batteries are made in sweat-shops in China, India and Korea, and I doubt you can find anybody to sue. So play it safe, both for your own sake and the camera's sake. If you want a spare, get a real Canon one.</td></tr></table>"
},
"TABREF2": {
"html": null,
"type_str": "table",
"text": "",
"num": null,
"content": "<table><tr><td>Annota-</td><td/><td colspan=\"3\">Annotation 2</td><td/></tr><tr><td>tion 1</td><td colspan=\"3\">best good fair</td><td>bad</td><td>total</td></tr><tr><td>best</td><td>294</td><td>44</td><td>2</td><td>0</td><td>340</td></tr><tr><td>good</td><td>66</td><td>639</td><td>113</td><td>0</td><td>818</td></tr><tr><td>fair</td><td>0</td><td colspan=\"4\">200 1,472 113 1,785</td></tr><tr><td>bad</td><td>1</td><td>2</td><td>78</td><td colspan=\"2\">1,885 1,966</td></tr><tr><td>total</td><td>361</td><td colspan=\"4\">885 1,665 1,998 4,909</td></tr></table>"
},
"TABREF3": {
"html": null,
"type_str": "table",
"text": "",
"num": null,
"content": "<table><tr><td>Separators</td></tr></table>"
},
"TABREF6": {
"html": null,
"type_str": "table",
"text": "",
"num": null,
"content": "<table><tr><td/><td>2% 4% 4% 10% 14% 22%</td></tr><tr><td colspan=\"2\">Battery 10% 14% 18% 30% 38% 50%</td></tr><tr><td colspan=\"2\">LCD 12% 18% 20% 22% 24% 28%</td></tr><tr><td>Flash</td><td>6% 10% 16% 20% 26% 42%</td></tr><tr><td>MM</td><td>6% 8% 8% 12% 18% 26%</td></tr><tr><td>%Query</td><td>RatioOfChange (-) <-0.30 <-0.25 <-0.20 <-0.15 <-0.10 <-0.05</td></tr><tr><td>IQ</td><td>4% 6% 10% 14% 18% 44%</td></tr><tr><td colspan=\"2\">Battery 2% 4% 4% 10% 14% 22%</td></tr><tr><td>LCD</td><td>4% 4% 8% 12% 22% 28%</td></tr><tr><td>Flash</td><td>4% 6% 8% 16% 18% 28%</td></tr><tr><td>MM</td><td>8% 10% 16% 18% 34% 42%</td></tr></table>"
}
}
}
} |