Research On Target-dependent Sentiment Analysis Of User Reviews

With the rapid development of Internet technology,the relationship between the Internet and people's lives is becoming more and more intensive.People create,share,and distribute information on the web,and there is a huge amount of information produced on the Internet every day.Among them,there are many user reviews.Analysis and mining of these information are of great significance to businesses and social organizations.Based on this background,this paper studies the target-dependent sentiment analysis of user reviews.This research aims to extract the aspect term of good or service involved in the user reviews,and to analyze the sentiment expressed by users on every aspect term.The main contributions of our work can be summarized as follows:1.The research of this part is to obtain aspect term in user reviews.To extract aspect term,this paper improved the rule based and crf based aspect term extraction algorithm.Then this paper designed and implemented BiLSTM based aspect term extraction model,and we utilized Tri-Letter to introduce character level text features,so that the model can gain more morphological information.2.The research of this part is to analyse the sentiment polarity of certain target.Traditional attention model can't fully capture long-range information.To solve this problem,this paper proposed DAT-LSTM,which caprured long-range informaton by introducing the dependency information.And we proposed Seg-DAT-LSTM,which segregated sentences into two shorter parts by the position of target word and respectively processed them,thus to fully capture long-range information.In order to make full use of the dependency context information,we proposed DPEAT-LSTM,combining dependency path information.