ASPECT LEVEL SENTIMENT CLASSIFICATION WITH DUAL POSITION AWARENESS BASED ON GRAPH-LSTMS
At present,most of the researches on the aspect term sentiment classification task in user reviews ignore the dependency syntactic information,or do not establish the relationship between dependency syntactic structure and words.Therefore,this paper proposes an aspect level sentiment classification method based on Graph-LSTMs.Graph-LSTMs was used to learn the context features of words.The position vector was spliced with dual position information in the input of bidirectional GRU to optimize the sentence sentiment coding.The attention mechanism was used to capture key sentiment features to achieve classification.The experimental results on two data sets of SemEval2014 show that the accuracy and Macro-F1 of this model are significantly improved compared with several baseline models.
国家科技期刊平台
NSTL
万方数据
维普
