Research Of Sentiment Analysis Based On Attention Bi-LSTM

With the rapid development of Internet technology,the breadth and depth of people’s participation in network behavior has been expanding.Therefore,a large amount of information data is generated on the Internet In the era of big data with information explosion,the potential value of data is huge.How to mine and use it from massive data has become the focus of attention of all walks of life.Text data occupies an important position in the mass data on the Internet.Not only is the amount of data huge,but also the value is great.As a result,text analysis and natural language processing technologies have become research hotspots in academia and industry.As an important branch of natural language processing,sentiment analysis has important value and practical needs in the fields of marketing,fraud recognition,economic forecasting,and public opinion monitoring.This article focuses on the research of sentiment classification algorithms,and experiments are performed with Douban movie reviews as a data set.The main work of this article is as follows:(1)This paper combines the Word2Vec model,the Bi-LSTM network,and the Attention mechanism to propose a hybrid model for Chinese sentiment analysis tasks,namely the Attention BiLSTM model.Firstly,the Word2Vec model is used to convert the preprocessed text data into semantic word vectors,then input the word vectors into the Bi-LSTM network for training,and finally input the output of the Bi-LSTM network into the Attention model and connect to the Softmax layer to give sentiment classification.The results show that the Attention Bi-LSTM model performs better than traditional machine learning models and deep learning models on sentiment analysis tasks.(2)This paper enhances the Attention mechanism based on the Attention Bi-LSTM model,and proposes an improved Attention Bi-LSTM model.By constructing a two-layer attention mechanism,first construct a word-level attention layer and then a sentence-level attention layer to better obtain the emotional information in the text and achieve a more accurate classification effect.Experimental results show that the improved Attention mechanism performs better than the ordinary Attention mechanism.