基于广义Jaccard系数的微博情感新词判定

微博情感新词的极性判定是情感分析研究中的一项基本任务,旨在对新词进行情感分类。针对极性判定的问题,提出一种新的计算特征向量相似度的算法。该方法首先使用特征向量表示情感新词和已有情感词,利用点互信息计算特征权值:然后采用广义Jaccard系数分别计算情感新词与已有的三种极性的情感词集内情感词的相似度,词集内相似度之和即为情感新词与该情感词集的相关度:最后,通过情感新词与三个极性情感词集的相关度的距离差判定其极性。实验结果表明,基于广义Jaccard系数的情感新词极性判定算法得出的F值比COAE 2014参赛队伍的最好成绩高两个百分点。

New microblog sentiment lexicon judgment based on generalized Jaccard coefficient

New microblog sentiment lexicon polarity judgment is a basic task aiming at classifying its emotion categories in sentiment analysis. This paper proposed a new approach that can judge the polarity of new microblog sentiment lexicon. The feature vectors are employed to represent new sentiment lexicon and the existing sentiment lexicon while the weight values are calculated by PMI. The similarity between the new sentiment lexicon and the candidates which is from three sentiment lexicon sets of different polarities through the generalized Jaccard coefficient, and the relativity between the new sentiment lexicon and the existing sentiment lexicon sets is defined as the sum of the above similarities. Finally, relativity distance differences of the three sentiment lexicon sets are applied to judge the polarity. The result of experiment showed that the F-score calculated through polarity judgment algorithm base on the generalized Jaccard coefficient was two points higher than the best team in COAE 2014.