文章建立了一套不受領域約束、在各個領域中通用的情感評價單元抽取模式,實現自動且準確地抽取情感評價單元。首先,基于短語的內部結構和句法功能,深入地分析評價對象及其對應的評價短語在句中的句法位置;其次,結合情感句中詞性和詞對間的依存關系進行情感評價單元的抽取;最后,通過手工構建通用和領域情感詞典進行情感評價單元抽取和情感傾向性計算的對比實驗,進一步驗證抽取模式的有效性。結合短語句法結構和依存句法進行情感評價單元抽取,準確地挖掘出評價對象及其對應的評價短語之間的句法關系。實驗結果表明,在應用到后續情感傾向性計算的實驗時,整體準確率得到大幅度提升。

The purpose of this paper is to establish a general appraisal expression extracting mode without domain-specific constraint, and realize appraisal expression extraction automatically and accurately. Firstly, the paper fully analyzes the syntactic position of appraisal subjects and their corresponding phrase based on the inner structure and syntactic function of the phrases. Then combining with dependency relationship between the part-of-speech and word pair, the paper extracts the appraisal expression. Finally, through manual construction of general sentiment lexicon and domain sentiment lexicon, the paper conducts compared experiments on appraisal expression extraction and emotion tendentiousness computing to test the effectiveness of the extraction mode. By combining syntactic structure of phrase and dependency parsing to extract appraisal expression, the paper could mine the syntactic relation between appraisal subjects and their corresponding phrase accurately. The experimental result shows when it applies the extracted appraisal expression to the sentiment computing, the whole accuracy is improved greatly.