File size: 70,005 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 | {
"paper_id": "R09-1010",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T15:00:07.816438Z"
},
"title": "Cross-Linguistic Sentiment Analysis: From English to Spanish",
"authors": [
{
"first": "Julian",
"middle": [],
"last": "Brooke",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Simon Fraser University Burnaby",
"location": {
"region": "BC",
"country": "Canada"
}
},
"email": ""
},
{
"first": "Milan",
"middle": [],
"last": "Tofiloski",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Simon Fraser University Burnaby",
"location": {
"region": "BC",
"country": "Canada"
}
},
"email": ""
},
{
"first": "Maite",
"middle": [],
"last": "Taboada",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Fraser University Burnaby",
"location": {
"region": "BC",
"country": "Canada"
}
},
"email": "mtaboada@sfu.ca"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We explore the adaptation of English resources and techniques for text sentiment analysis to a new language, Spanish. Our main focus is the modification of an existing English semantic orientation calculator and the building of dictionaries; however we also compare alternate approaches, including machine translation and Support Vector Machine classification. The results indicate that, although languageindependent methods provide a decent baseline performance, there is also a significant cost to automation, and thus the best path to long-term improvement is through the inclusion of language-specific knowledge and resources.",
"pdf_parse": {
"paper_id": "R09-1010",
"_pdf_hash": "",
"abstract": [
{
"text": "We explore the adaptation of English resources and techniques for text sentiment analysis to a new language, Spanish. Our main focus is the modification of an existing English semantic orientation calculator and the building of dictionaries; however we also compare alternate approaches, including machine translation and Support Vector Machine classification. The results indicate that, although languageindependent methods provide a decent baseline performance, there is also a significant cost to automation, and thus the best path to long-term improvement is through the inclusion of language-specific knowledge and resources.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Sentiment analysis refers to the automatic determination of subjectivity (whether a text is objective or subjective), polarity (positive or negative) and strength (strongly or weakly positive/negative). It is a growing field of research, especially given the gains to be obtained from mining opinions available online. Approaches to sentiment analysis have tackled the problem from two different angles: a word-based or semantic approach, or a machine learning (ML) approach. The word-based approach uses dictionaries of words tagged with their semantic orientation (SO), and calculates sentiment by aggregating the values of those present in a text or sentence [17] . The ML approach uses collections of texts that are known to express a favorable or unfavorable opinion as training data, and learns to recognize sentiment based on those examples [13] .",
"cite_spans": [
{
"start": 662,
"end": 666,
"text": "[17]",
"ref_id": "BIBREF16"
},
{
"start": 848,
"end": 852,
"text": "[13]",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Our approach is semantic, and makes use of a series of dictionaries, additionally taking into account the role of negation, intensification and irrealis expressions. We believe that a semantic approach offers the advantage of taking many different aspects of a text into account.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "One of the disadvantages of a semantic approach is that the resources necessary for a new domain or a new language need to be built from scratch, whereas a machine-learning approach only needs enough data to train. In this paper we show that porting to a new language, Spanish, requires only a small initial investment, while providing the opportunities for further improvement available only to semantic methods.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "For comparison, we have taken three approaches to performing sentiment analysis in a new language. Our main approach involves deploying Spanish-specific resources, which we build both manually and automatically. The second approach, used in Bautin et al. [4] and Wan [18] , consists of translating the texts into English, and using an existing English calculator. Finally, the third approach builds unigram Support Vector Machine classifiers from our Spanish corpora.",
"cite_spans": [
{
"start": 255,
"end": 258,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 267,
"end": 271,
"text": "[18]",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Our evaluation on multi-domain corpora indicates that, although translation and machine learning classification both perform reasonably well, there is a significant cost to automated translation. A languagespecific SO Calculator with dictionaries built using words that actually appear in relevant texts gives the best performance, with significant potential for improvement.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Our semantic orientation calculator (SO-CAL) uses five main dictionaries: four lexical dictionaries with 2,257 adjectives, 1,142 nouns, 903 verbs, and 745 adverbs, and a fifth dictionary containing 177 intensifying words and expressions. Although the vast majority of the entries are single words, our calculator also allows for multiword entries written in regular expression-like language.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "The SO-carrying words in these dictionaries were taken from a variety of sources, the three largest a corpus of 400 mixed reviews from Epinions.com, a 100 text subset of the 2,000 movie reviews in the Polarity Dataset [12] , and positive and negative words from the General Inquirer dictionary [15] . Each of the open-class words were given a hand-ranked SO value between 5 and -5 by a native English speaker. The numerical values were chosen to reflect both the prior polarity and strength of the word, averaged across likely interpretations. For example, the word phenomenal is a 5, nicely a 2, disgust a -3, and monstrosity a -5. The dictionary was later reviewed by a committee of three other researchers in order to minimize the subjectivity of ranking SO by hand.",
"cite_spans": [
{
"start": 218,
"end": 222,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 294,
"end": 298,
"text": "[15]",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "SO-CAL also implements a modified version of contextual valence shifting as originally proposed by Polanyi and Zaenen [14] , including negation and intensification. We have also added irrealis blocking.",
"cite_spans": [
{
"start": 118,
"end": 122,
"text": "[14]",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "Our approach to negation differs from Polanyi and Zaenen's in that negation involves a polarity shift instead of a switch: A negated adjective is shifted by a fixed amount (4) toward the origin. This means that the negation of a strongly negative word (like terrible) will be neutral or weakly negative (not terrible -5 + 4 = -1 instead of 5), while the negation of a weakly positive word like nice is equally negative (not nice 2 -4 = -2).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "The calculation of intensification is somewhat more sophisticated than simple addition and subtraction. Each expression in our intensifier dictionary is associated with a multiplier value. For instance, very has a value of .25, which means the SO value of any adjective modified by very is increased by 25%. We also included three other kinds of intensification that are common within our genre: the use of all capital letters, the use of exclamation points, and the use of discourse but to indicate more salient information (e.g., \u2026but the movie was GREAT!).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "Some markers indicate that the words appearing in a sentence might not be reliable for the purposes of sentiment analysis. We refer to these using the linguistic term irrealis. Irrealis markers in English include modals (would, could), some verbs (expect, doubt), and certain kinds of punctuation (questions, quotations). When SOcarrying words appear within the scope of these markers, our calculator ignores them.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "Lexicon-based sentiment classifiers generally show a positive bias [10] , likely the result of a human tendency to favor positive language [6] . In order to overcome this bias, we increase the final SO of any negative expression (after other modifiers have applied) by a fixed amount (currently 50%).",
"cite_spans": [
{
"start": 67,
"end": 71,
"text": "[10]",
"ref_id": "BIBREF9"
},
{
"start": 139,
"end": 142,
"text": "[6]",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "For initial testing, we use the 400 text Epinions corpus (50 texts in each of eight different product types), the other 1,900 texts in the Polarity Dataset (Movie), and a 2,400 text corpus of camera, printer, and stroller reviews (Camera) taken from a larger set of Epinions reviews also used by Bloom et al. [5] , for a total of 4,700 texts split equally between positive and negative. Table 1 shows the performance of the English calculator with all features, and disabling the three types of valence shifters (negation, intensification and irrealis) and the extra weight on negative words. An asterisk (*) indicates that a chi-square test yielded significance at the p<0.05 level, as compared to the result with all features enabled. Whereas not all the differences are statistically significant, it does seem that the set of features that we have chosen has a positive effect on performance. ",
"cite_spans": [
{
"start": 309,
"end": 312,
"text": "[5]",
"ref_id": "BIBREF4"
}
],
"ref_spans": [
{
"start": 387,
"end": 394,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "The English SO Calculator",
"sec_num": "2."
},
{
"text": "Compared to English, Spanish is a highly inflected language, with gender and plural markers on nouns, as well as a rich system of verbal inflection (45 possible verb forms). In the English version of SO-CAL, the only external software we made use of was the Brill tagger [7] ; lemmatization of noun and verbs was simple enough to be carried out during the calculation. For Spanish, we used a high-accuracy statistical tagger, the SVMTool [9] , and we adapted a 500,000+ word lemma dictionary included in the FreeLing software package 1 , which we used to both lemmatize the words and to add more detail to the basic verb tags assigned by SVMTool (each verb is lemmatized, but tagged with information about its tense and mood). We found that some sentiment-relevant words were not being lemmatized properly, so we also implemented a second layer of lemmatization within the calculator.",
"cite_spans": [
{
"start": 271,
"end": 274,
"text": "[7]",
"ref_id": "BIBREF6"
},
{
"start": 438,
"end": 441,
"text": "[9]",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "Most of the Python code written for the English version of SO-CAL could be reused. With regards to detecting negation, intensification, and modifier blocking, it was necessary to take into account the fact that in Spanish adjectives appear both before and (more commonly) after the noun. The most interesting difference was the fact that verb forms in Spanish provide irrealis information. In particular, the conditional tense and the imperative and subjunctive moods often serve to indicate that the situation being referred to is not in fact the case. Thus, in Spanish we used a mixture of word and inflection-based irrealis blocking, using the same words as the English version whenever possible.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "We built new Spanish dictionaries, including dictionaries for adjectives, nouns, verbs, adverbs and intensifiers. For intensifiers, given the fact that they are closed-class and highly idiosyncratic, we simply created a new list of 157 expressions, based on the English list. For the open-class dictionaries, we tested three different methods of dictionary-building; we compare their performance on the Spanish corpus in Section 5.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "The first set of dictionaries started with the English dictionaries for each part of speech, which we translated automatically into Spanish, preserving the semantic orientation value for each word. For the automatic translation we used, in turn, two different methods. The first was an online bilingual dictionary, from the site www.spanishdict.com. We extracted the first definition under the appropriate syntactic category, ignoring any cases where either the English or the Spanish were multiword expressions. The second automatic translation method involved simply plugging our English dictionaries into the Google translator and parsing the results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "For the second method of dictionary creation, we took the lists from Spanishdict.com and manually fixed entries that were obviously wrong. This involved mostly removing words in the wrong dictionary for their part of speech, but also changing some of the values (less than 10% for each dictionary). This hand-correction took a native speaker of Spanish about two hours to complete.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "Finally, the third method consisted in creating all dictionaries from scratch. Our source corpora created for this project consists of reviews extracted from the Ciao.es review website. Following the basic format of the Epinions corpus, we collected 400 reviews from the domains of hotels, movies, music, phones, washing machines, books, cars, and computers. Each category contained 50 reviews: 25 positive and 25 negative. Whenever possible, exactly two reviews, one positive and one negative, were taken for any particular product, so that the machine learning classifier described in Section 4.2 could not use names as sentiment clues.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "We tagged the Spanish corpus collected from Ciao.es, and extracted all adjectives, nouns, adverbs and verbs. This resulted in large lists for each category (e.g., over 10,000 nouns). We manually pruned the lists, removing words that did not convey sentiment, misspelled and inflected words, and words with the wrong part of speech tag. Finally, semantic orientation values were assigned for each. This process took a native speaker of Spanish about 12 hours. We decided against a committee review of the Spanish dictionaries for the time being.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "Another type of dictionary tested was a merging of the dictionaries created using the second and third methods, i.e., the automatically-created (but hand-fixed) dictionaries and the ones created from scratch (Ciao manual). We created two versions of these dictionaries, depending on whether we used the value from the Fixed Spanishdict.com or Ciao dictionary. The dictionaries range from smallest (Spanishdict.com) to largest (Ciao+Fixed). The first one contains 1,160 adjectives, 979 nouns, 500 verbs and 422 adverbs. The combined dictionary has 2,049 adjectives, 1,324 nouns, 739 verbs, and 548 adverbs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "We performed a comparison of fully automated and fully manual methods, comparing the unedited Spanishdict.com dictionaries and the ones created by hand. We calculated the percentage of words in common, as a percentage of the size for the larger of the two sets (the Spanishdict.com dictionaries). The commonalities ranged from roughly 20% of the words for nouns to 41% for adjectives (i.e., 41%, or 480 of the hand-ranked adjectives were also found in the automatic dictionary). We also compared the values assigned to each word: The variance of the error ranged from 1.001 (verbs) to 1.518 (adjectives). Automatically translated dictionaries tend to include more formal words, whereas the ones created by hand include many more informal and slang words 4. Alternative approaches",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Spanish SO Calculator",
"sec_num": "3."
},
{
"text": "For translation, we used Google's web-based translation system. Google Translate (translate.google.com) uses phrase-based statistical machine translation. We used only one translator, but Bautin et al. [4] discuss the use of different Spanish translating systems, and Wan [18] compare Chinese machine translators; the latter found that Google gave the best performance, which is consistent with our preliminary testing of other systems.",
"cite_spans": [
{
"start": 202,
"end": 205,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 272,
"end": 276,
"text": "[18]",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Corpus translation",
"sec_num": "4.1"
},
{
"text": "A popular approach to sentiment analysis has been the automatic training of a text classifier. Cross-linguistic sentiment detection seems particularly amenable to machine learning, since classifiers can be easily trained in any language. Following Pang et al. [13] , we used Support Vector Machine (SVM) classifiers, built with the sequential minimal optimization algorithm included in the WEKA software suite [20] , with a linear kernel and testing done with 10-fold cross-validation. We trained using unigram features that appeared at least four times in the dataset (the same cut-off was used by Pang and Lee [12] ). To test the efficacy of the WEKA classifiers, we first trained a classifier on the full 2,000 text Polarity Dataset, a collection of balanced positive and negative movie reviews [12] , comparing the cross-validated results with the baseline for SVM unigram classifiers on this dataset (before other improvements) given in Pang and Lee [12] . The difference (about 1%) was not statistically significant. It is worth noting that more recent work in SVM-based sentiment analysis has shown significant improvement on this baseline [19] , however relevant resources are not available for Spanish.",
"cite_spans": [
{
"start": 260,
"end": 264,
"text": "[13]",
"ref_id": "BIBREF12"
},
{
"start": 410,
"end": 414,
"text": "[20]",
"ref_id": "BIBREF19"
},
{
"start": 612,
"end": 616,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 798,
"end": 802,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 955,
"end": 959,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 1147,
"end": 1151,
"text": "[19]",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Machine Learning",
"sec_num": "4.2"
},
{
"text": "In order to compare the classifier across languages, we trained separately on each of our two 400-text development corpora. In each case we used the output after pre-processing, with lemmatizing in the case of Spanish. In addition to basic unigrams we also tested unigrams with full POS tags and, for Spanish, partial tags (retaining word class but disregarding inflection such as number and person). The results were identical or in some cases worse than a simple unigram model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine Learning",
"sec_num": "4.2"
},
{
"text": "We built two additional 400 text corpora, in English and Spanish, with the same basic constituency as the Epinions and Ciao Corpus discussed earlier. The English corpus (Epinions 2) is also from the Epinions site, while the Spanish corpus came from Dooyoo.es. This second set of texts for each language has never been used for training or development of any of our resources All four corpora were translated using the appropriate Google translator, and for each version the accuracy identifying the polarity of reviews for all possible dictionaries and methods was tested. Note that when the corpus and the dictionary are the same language, the original version of the corpus is used, and when the corpus and the dictionary are in different languages, we use the translated version. The results are given in Table 2 .",
"cite_spans": [],
"ref_spans": [
{
"start": 808,
"end": 815,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5."
},
{
"text": "There are a number of clear patterns in Table 2 . First, for the original Spanish versions, the translated Spanish dictionaries, taken together, do poorly compared to the versions of the dictionaries derived from actual Spanish texts; this is significant at the p<0.05 level for all possible dictionary combinations (all significance results are derived from chi-square tests). For Spanish, including words from translated dictionaries has little or no benefit. The opposite is true for Spanish translations of English texts, where the Ciao (manual) dictionary performance is low, and performance improves dramatically with the addition of translated (although manually fixed) resources; in the case of the Epinions 2 corpus, this improvement is significant (p<0.05). We attribute this to the fact that translated texts and translated dictionaries \"speak the same language\"; translated English corpora are unlikely to contain the colloquial Spanish found in the Ciao dictionary, and are more likely to contain the kind of formal language we saw in our translated dictionaries. Turning now to machine learning methods, the SVM classifiers show the worse performance overall, however only the difference seen in the Epinions 2 corpus is significant (at the p<0.01 level). The relatively poor performance of the SVM classifier in this case can be attributed to the small size of the training set and the heterogeneity of the corpora; SVM classifiers have been shown to have poor cross-domain performance in text sentiment tasks [2] , a problem that can be remedied somewhat by integrating a lexicon-based system [1] .",
"cite_spans": [
{
"start": 1525,
"end": 1528,
"text": "[2]",
"ref_id": "BIBREF1"
},
{
"start": 1609,
"end": 1612,
"text": "[1]",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 40,
"end": 47,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5."
},
{
"text": "The numbers in Table 2 do not indicate a clear winner with respect to the performance of Spanish SO-CAL as compared to English SO-CAL with translated texts, although it is clear that translating English texts into Spanish is, at present, a bad approach (p<0.01). The totals for all corpora for each method suggest that Spanish SO-CAL is performing well below English SO-CAL (p<0.01). Table 3 summarizes the effects of translation. Original refers to all the 1,600 original versions and Translated to all 1,600 translated versions. For SO calculation, we use the best performing dictionary in the relevant language. Table 3 shows a general deficit for translated texts; for SO calculation, this is significant at the p<0.01 level. The fact that it is also visible in SVMs (which are not subject to dictionary biases) suggests that it is a general phenomenon. One potential criticism here is our use of corpora whose words were the basis for our dictionary, unfairly providing two of the four original corpora with high coverage which would not pass to the translations. Indeed, there is some evidence in Table 3 to suggest that these high coverage corpora do outperform their low coverage counterparts to some degree in relevant dictionaries (compared with the Subjective dictionary, for instance); in general, though, there were no significant differences among same-language corpora tested using the same dictionary. Note also that using high-coverage corpora is not analogous to testing and training on the same corpora, since words are rated for SO independently of the texts in which they appear.",
"cite_spans": [],
"ref_spans": [
{
"start": 15,
"end": 22,
"text": "Table 2",
"ref_id": "TABREF1"
},
{
"start": 384,
"end": 391,
"text": "Table 3",
"ref_id": "TABREF2"
},
{
"start": 615,
"end": 622,
"text": "Table 3",
"ref_id": "TABREF2"
},
{
"start": 1103,
"end": 1110,
"text": "Table 3",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5."
},
{
"text": "Wan [18] created a hybrid classifier which combined the scores from a Chinese lexicon-based system and an English lexicon-based system (with translated texts). In contrast to our results, his Chinese lexicon-based system performed quite poorly compared to the English system. Similar to our results, Chinese lexicons created by translating English lexicons did not help performance.",
"cite_spans": [
{
"start": 4,
"end": 8,
"text": "[18]",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "6."
},
{
"text": "Although they are concerned with sentence level subjectivity instead of text-level polarity, the work of Mihalcea et al. [11] is quite relevant, since their focus, like ours, is on exploring ways to deriving new resources from existing resources for English. In adapting subjectivity cues to Romanian, they also saw limited benefits to straight translation of dictionaries, but obtained promising results from the projection of English annotations into Romanian.",
"cite_spans": [
{
"start": 121,
"end": 125,
"text": "[11]",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "6."
},
{
"text": "Bautin et al. [4] used online resources from multiple languages, including Spanish, into English, using the output from an existing sentiment analyzer to track attitudes in different language communities. Yao et al. [21] made use of a bilingual lexicon to build a Chinese sentiment dictionary using English glosses. Lexiconbased sentiment analysis has also been pursued independently in a number of East Asian languages, including Japanese [16] , Chinese [22] , and Korean [8] . As far as we know, ours is the first Spanish SO calculator. Banea et al. [3] report on work in Spanish, but theirs is a subjectivity classification task.",
"cite_spans": [
{
"start": 14,
"end": 17,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 216,
"end": 220,
"text": "[21]",
"ref_id": "BIBREF20"
},
{
"start": 440,
"end": 444,
"text": "[16]",
"ref_id": "BIBREF15"
},
{
"start": 455,
"end": 459,
"text": "[22]",
"ref_id": "BIBREF21"
},
{
"start": 473,
"end": 476,
"text": "[8]",
"ref_id": "BIBREF7"
},
{
"start": 552,
"end": 555,
"text": "[3]",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "6."
},
{
"text": "In terms of approaches to calculation of text level sentiment in English, the work of Kennedy and Inkpen [10] is the most directly comparable. Their main focus was the comparison of lexicon-based versus machine learning approaches; in contrast to our results, they found that performance of their semantic model was significantly below that of an SVM classifier.",
"cite_spans": [
{
"start": 105,
"end": 109,
"text": "[10]",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "6."
},
{
"text": "To facilitate comparisons with other approaches, the corpora and some of the resources described in the paper are available 2 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "6."
},
{
"text": "The surge in attention paid to automated analysis of text sentiment has largely been focused on English. In this paper, we have discussed how to adapt an existing English semantic orientation system to Spanish while at the same time comparing several alternative approaches.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7."
},
{
"text": "Our results indicate that SVMs, at least the fairly simple SVMs we have tested here, do not do very well in our Spanish corpora. There are a number of obvious reasons for this, and our rejection of SVMs is far from decisive; on the contrary, machine learning might be useful, for instance, in identifying parts of the text that should be disregarded during the SO calculation [12] .",
"cite_spans": [
{
"start": 376,
"end": 380,
"text": "[12]",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7."
},
{
"text": "For calculation of semantic orientation using lexicons, translation of any kind seems to come with a price, even between closely related languages such as English and Spanish. Our Spanish SO calculator (SO-CAL) is clearly inferior to our English SO-CAL, probably the result of a number of factors, including a small, preliminary dictionary, and a need for additional adaptation to a new language. Translating our English dictionary also seems to result in significant semantic loss, at least for original Spanish texts. Although performance of Spanish texts translated into English is comparable to native SO-CAL performance, the overall accuracy of translated texts in both English and Spanish suggests that there is 3-5% performance cost for any (automated) translation. This, together with the fact that translation seems to have a disruptive effect on previous reliable improvements, as well as the relatively small time investment required to develop Spanish SO-CAL, lead us to conclude that there is value in pursuing the development of language-specific resources, notwithstanding new breakthroughs in machine translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7."
},
{
"text": "http://garraf.epsevg.upc.es/freeling/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work was supported by a NSERC Discovery Grant (261104-2008) to Maite Taboada.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": "8."
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "When specialists and generalists work together: Domain dependence in sentiment tagging",
"authors": [
{
"first": "A",
"middle": [],
"last": "Andreevskaia",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Bergler",
"suffix": ""
}
],
"year": 2008,
"venue": "Proc. of 46th ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Andreevskaia and S. Bergler. When specialists and generalists work together: Domain dependence in sentiment tagging. Proc. of 46th ACL. Columbus, OH, 2008.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Customizing sentiment classifiers to new domains: A case study",
"authors": [
{
"first": "A",
"middle": [],
"last": "Aue",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Gamon",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of RANLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Aue and M. Gamon. Customizing sentiment classifiers to new domains: A case study. Proc. of RANLP. Borovets, Bulgaria, 2005.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Multilingual subjectivity analysis using machine translation",
"authors": [
{
"first": "C",
"middle": [],
"last": "Banea",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Wiebe",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Hassan",
"suffix": ""
}
],
"year": 2008,
"venue": "Proc. of EMNLP. Honolulu",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "C. Banea, R. Mihalcea, J. Wiebe and S. Hassan. Multilingual subjectivity analysis using machine translation. Proc. of EMNLP. Honolulu, 2008.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "International sentiment analysis for news and blogs",
"authors": [
{
"first": "M",
"middle": [],
"last": "Bautin",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Vijayarenu",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Skiena",
"suffix": ""
}
],
"year": 2008,
"venue": "Proc. of 3rd AAAI International Conference on Weblogs and Social Media",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Bautin, L. Vijayarenu and S. Skiena. International sentiment analysis for news and blogs. Proc. of 3rd AAAI International Conference on Weblogs and Social Media. San Jose, CA, 2008.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Extracting appraisal expressions",
"authors": [
{
"first": "K",
"middle": [],
"last": "Bloom",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Navendu",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Argamon",
"suffix": ""
}
],
"year": 2007,
"venue": "Proc. of HLT/NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "K. Bloom, G. Navendu and S. Argamon. Extracting appraisal expressions. Proc. of HLT/NAACL. Rochester, NY, 2007.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "The Pollyanna hypothesis",
"authors": [
{
"first": "J",
"middle": [
"D"
],
"last": "Boucher",
"suffix": ""
},
{
"first": "C",
"middle": [
"E"
],
"last": "Osgood",
"suffix": ""
}
],
"year": 1969,
"venue": "Journal of Verbal Learning and Verbal Behaviour",
"volume": "8",
"issue": "",
"pages": "1--8",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J.D. Boucher and C.E. Osgood. The Pollyanna hypothesis. Journal of Verbal Learning and Verbal Behaviour 8: 1-8, 1969.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "A simple rule-based part of speech tagger",
"authors": [
{
"first": "E",
"middle": [],
"last": "Brill",
"suffix": ""
}
],
"year": 1992,
"venue": "Proc. of 3rd Conference on Applied Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "E. Brill. A simple rule-based part of speech tagger. Proc. of 3rd Conference on Applied Natural Language Processing. Trento, Italy, 1992.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Automatic affect recognition using natural language processing techniques and manually built affect lexicon",
"authors": [
{
"first": "Y",
"middle": [
"H"
],
"last": "Cho",
"suffix": ""
},
{
"first": "K",
"middle": [
"J"
],
"last": "Lee",
"suffix": ""
}
],
"year": 2006,
"venue": "IEICE Transactions on Information and Systems",
"volume": "89",
"issue": "12",
"pages": "2964--2971",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y.H. Cho and K.J. Lee. Automatic affect recognition using natural language processing techniques and manually built affect lexicon. IEICE Transactions on Information and Systems 89(12): 2964-2971, 2006.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "SVMTool: A general POS tagger generator based on support vector machines",
"authors": [
{
"first": "J",
"middle": [],
"last": "Gim\u00e9nez",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "M\u00e0rquez",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of Conference on Language Resources and Evaluation (LREC)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Gim\u00e9nez and L. M\u00e0rquez. SVMTool: A general POS tagger generator based on support vector machines. Proc. of Conference on Language Resources and Evaluation (LREC). Lisbon, Portugal, 2004.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Sentiment classification of movie and product reviews using contextual valence shifters",
"authors": [
{
"first": "A",
"middle": [],
"last": "Kennedy",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Inkpen",
"suffix": ""
}
],
"year": 2006,
"venue": "Computational Intelligence",
"volume": "22",
"issue": "2",
"pages": "110--125",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Kennedy and D. Inkpen. Sentiment classification of movie and product reviews using contextual valence shifters. Computational Intelligence 22(2): 110-125, 2006.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Learning multilingual subjective language via cross-lingual projections",
"authors": [
{
"first": "R",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Banea",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Wiebe",
"suffix": ""
}
],
"year": 2007,
"venue": "Proc. of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Mihalcea, C. Banea and J. Wiebe. Learning multilingual subjective language via cross-lingual projections. Proc. of ACL. Prague, Czech Republic, 2007.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts",
"authors": [
{
"first": "B",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "B. Pang and L. Lee. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts. Proc. of ACL. Barcelona, Spain, 2004.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Thumbs up? Sentiment classification using Machine Learning techniques",
"authors": [
{
"first": "B",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Vaithyanathan",
"suffix": ""
}
],
"year": 2002,
"venue": "Proc. of EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "B. Pang, L. Lee and S. Vaithyanathan. Thumbs up? Sentiment classification using Machine Learning techniques. Proc. of EMNLP, 2002.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Contextual valence shifters",
"authors": [
{
"first": "L",
"middle": [],
"last": "Polanyi",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Zaenen",
"suffix": ""
}
],
"year": 2006,
"venue": "Computing Attitude and Affect in Text: Theory and Applications",
"volume": "",
"issue": "",
"pages": "1--10",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "L. Polanyi and A. Zaenen. Contextual valence shifters. In Computing Attitude and Affect in Text: Theory and Applications, J.G. Shanahan, Y. Qu, and J. Wiebe, Eds. Springer: Dordrecht, pp. 1-10, 2006.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Thematic text analysis: New agendas for analyzing text content",
"authors": [
{
"first": "P",
"middle": [
"J"
],
"last": "Stone",
"suffix": ""
}
],
"year": 1997,
"venue": "Text Analysis for the Social Sciences",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P.J. Stone. Thematic text analysis: New agendas for analyzing text content. In Text Analysis for the Social Sciences, C. Roberts, Ed. Lawrence Erlbaum: Mahwah, NJ, 1997.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Extracting semantic orientations of words using spin model",
"authors": [
{
"first": "H",
"middle": [],
"last": "Takamura",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Inui",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Okumura",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "H. Takamura, T. Inui and M. Okumura. Extracting semantic orientations of words using spin model. Proc. of ACL. Ann Arbor, 2005.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews",
"authors": [
{
"first": "P",
"middle": [],
"last": "Turney",
"suffix": ""
}
],
"year": 2002,
"venue": "Proc. of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. Turney. Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews. Proc. of ACL, 2002.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Using bilingual knowledge and ensemble techniques for unsupervised Chinese sentiment analysis",
"authors": [
{
"first": "X",
"middle": [],
"last": "Wan",
"suffix": ""
}
],
"year": 2008,
"venue": "Proc of EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "X. Wan. Using bilingual knowledge and ensemble techniques for unsupervised Chinese sentiment analysis. Proc of EMNLP. Honolulu, 2008.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Using Appraisal groups for sentiment analysis",
"authors": [
{
"first": "C",
"middle": [],
"last": "Whitelaw",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Garg",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Argamon",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of ACM SIGIR Conference on Information and Knowledge Management (CIKM 2005)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "C. Whitelaw, N. Garg and S. Argamon. Using Appraisal groups for sentiment analysis. Proc. of ACM SIGIR Conference on Information and Knowledge Management (CIKM 2005). Bremen, Germany, 2005.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Data Mining: Practical Machine Learning Tools and Techniques",
"authors": [
{
"first": "I",
"middle": [
"H"
],
"last": "Witten",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2005,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "I.H. Witten and E. Frank. Data Mining: Practical Machine Learning Tools and Techniques. 2nd ed. San Francisco: Morgan Kaufmann, 2005.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Using bilingual lexicon to judge sentiment orientation of Chinese words",
"authors": [
{
"first": "J",
"middle": [],
"last": "Yao",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Zheng",
"suffix": ""
}
],
"year": 2006,
"venue": "Proc. of 6th International Conference on Computer and Information Technology (CIT'06)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Yao, G. Wu, J. Liu and Y. Zheng. Using bilingual lexicon to judge sentiment orientation of Chinese words. Proc. of 6th International Conference on Computer and Information Technology (CIT'06). Seoul, Korea, 2006.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Sentiment classification for Chinese reviews: A comparison between SVM and semantic approaches",
"authors": [
{
"first": "Q",
"middle": [],
"last": "Ye",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Y.-J",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2005,
"venue": "Fourth Int. Conference on Machine Learning and Cybernetics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Q. Ye, B. Lin and Y.-J. Li. Sentiment classification for Chinese reviews: A comparison between SVM and semantic approaches. Fourth Int. Conference on Machine Learning and Cybernetics. Guangzhou, China, 2005.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"content": "<table><tr><td>All</td><td>80.3</td><td>76.4</td><td>80.3</td><td>78.7*</td></tr><tr><td>No Neg</td><td>75.8*</td><td>74.6</td><td>76.1*</td><td>75.4*</td></tr><tr><td>No Int</td><td>79.0*</td><td>74.7</td><td>77.5*</td><td>76.5*</td></tr><tr><td>No Irreal</td><td>78.8*</td><td>74.8</td><td>79.6</td><td>77.6*</td></tr><tr><td>No Neg W</td><td>71.8*</td><td>75.6</td><td>71.5*</td><td>73.2*</td></tr></table>",
"type_str": "table",
"html": null,
"text": "",
"num": null
},
"TABREF1": {
"content": "<table><tr><td>Corpus</td></tr></table>",
"type_str": "table",
"html": null,
"text": "",
"num": null
},
"TABREF2": {
"content": "<table><tr><td>Method</td><td>Texts</td><td>Accuracy</td></tr><tr><td>SO Calculation</td><td>Original Translated</td><td>76.62 71.81</td></tr><tr><td>SVM</td><td>Original Translated</td><td>72.56 69.25</td></tr></table>",
"type_str": "table",
"html": null,
"text": "",
"num": null
}
}
}
} |