电子科技大学学报
電子科技大學學報
전자과기대학학보
JOURNAL OF UNIVERSITY OF ELECTRONIC SCIENCE AND TECHNOLOGY OF CHINA
2014年
3期
420-425
,共6页
张磊%李珊%彭舰%陈黎%黎红友
張磊%李珊%彭艦%陳黎%黎紅友
장뢰%리산%팽함%진려%려홍우
依存关系%特征-情感对%特征模板%最大熵%Web数据挖掘
依存關繫%特徵-情感對%特徵模闆%最大熵%Web數據挖掘
의존관계%특정-정감대%특정모판%최대적%Web수거알굴
dependency relations%feature-opinion pairs%feature template%maximum entropy model%Web data mining
中文产品评论特征词与关联的情感词的分类是观点挖掘的重要研究内容之一。该文改进了英文依存关系语法,总结出5种常用的中文产品评论依存关系;利用最大熵模型进行训练,设计了基于依存关系的复合特征模板。实验证明,应用该复合模板进行特征-情感对的提取,系统的查全率和F-score相比于传统方法,分别提高到78.68%和75.36%。
中文產品評論特徵詞與關聯的情感詞的分類是觀點挖掘的重要研究內容之一。該文改進瞭英文依存關繫語法,總結齣5種常用的中文產品評論依存關繫;利用最大熵模型進行訓練,設計瞭基于依存關繫的複閤特徵模闆。實驗證明,應用該複閤模闆進行特徵-情感對的提取,繫統的查全率和F-score相比于傳統方法,分彆提高到78.68%和75.36%。
중문산품평론특정사여관련적정감사적분류시관점알굴적중요연구내용지일。해문개진료영문의존관계어법,총결출5충상용적중문산품평론의존관계;이용최대적모형진행훈련,설계료기우의존관계적복합특정모판。실험증명,응용해복합모판진행특정-정감대적제취,계통적사전솔화F-score상비우전통방법,분별제고도78.68%화75.36%。
In recent years, feature-opinion pairs classification of Chinese product review is one of the most important research field in Web data mining technology. In this paper, five types of Chinese dependency relationships for product review have been concluded based on the traditional English dependency grammar. The maximum entropy model is used to predict the opinion-relevant product feature relations. To train the model, a set of feature symbol combinations have been designed by means of Chinese dependency. The experiment result shows that the recall and F-score of our approach could reach 78.68%and 75.36%respectively, which is clearly superior to Hu’s adjacent based method and Popesecu’s pattern based method.