Research On Aspect-based Sentiment Analysis Based On Attention Mechanism

Sentiment analysis is an important task in NLP.It has a significance impact on modern life with user feedback mining.Text sentiment analysis satisfies the needs of public opinion analysis,product demand,and opinion mining by summarizing the user emotions behind specific objects.Aspect based sentiment analysis is an important sub-task of sentiment analysis,and it is also the core issue of the semantic web and computational linguistics in recent years.It aims to predict the sentiment polarity of the given aspect term in the sentence.In view of the problem,this paper proposes a position-aware multi-head attention network for aspect based sentiment analysis(PMANet)and a word embedding fusion interactive model for sentiment analysis(WE-fusion).The main tasks as follows:(1)In order to improve the performance of sentiment analysis,we proposed a position-aware multi-head attention network for aspect based sentiment analysis(PMANet),which the combination of the GRU network and multi-head attention mechanism plays an effective role.On the one hand,the model extracts time series information with GRU network.On the other hand,the model extracts the vertical space feature information with multi-head attention mechanism.The model retains the position information of the aspect term,which could improve the information retrieval performance.Whereas,the interaction attention mechanism can learn the semantics between the aspect term and context.(2)In order to incorporate more knowledge in the word embedding,we proposes a word embedding fusion interactive model for sentiment analysis(WE-fusion).RWE encodes relationship knowledge in word embedding,and FastText take into account the effect of sentence sequences into the word embedding.Therefore,the hidden state representation obtained by the GloVe and the RWE(or FastText)with Bi-LSTM network,which computing the weight distribution by attention mechanism.Whereas the interactive attention mechanism learn the relationship between the aspect term and context,so as to achieve the effect of improving the prediction performance of model.(3)Finally,with the experiments on the public data set SemEval 2014,it proves the effective of our model that proposed in this paper.