A Word2Vec,LSTMs And Attention Based Model For Chinese Sentiment Analysis

With the rapid development of Web2.0 technology,our internet society has entered the era of big data.Nowadays,the value of data has been widely recognized,and mining valuable information from massive data has become a new profit growth point for all walks of life.Traditional data mining mainly focuses on structured data,recently,research and application of unstructured data have also received increasing attention.As one of the richest source in the internet,text data contains a huge amount of valuable information,therefore,text analysis and natural language processing(NLP)technology has gradually become a research hot spot in academia and industry.Sentiment analysis is an important branch of natural language processing,it has been widely used in many fields such as public opinion monitoring,marketing,fraud detection and economic forecasting.Due to the differences in language features,Chinese is more complex and difficult to deal with in natural language processing than English.This paper focuses on algorithms for Chinese sentiment classification,and selects douban movie short review data set and multi-category commodity review data set for experiments.Compared with traditional machine learning algorithms,deep learning algorithms reduce the trouble of manually extracting features and have better performance in many tasks,deep learning has been widely used in various fields of artificial intelligence.In this paper,based on Word2 Vec model,LSTMs network and Attention mechanism,a hybrid framework Word2Vec-LSTMs-Attention is proposed for Chinese sentiment classification.The main steps of the model are: 1.Word2 Vec model is used to transform Chinese words into finite-dimensional real number vectors;2.The word vectors are input into LSTMs network for training;3.Outputs of the LSTMs layer serve as the input of the attention layer,and finally connects to the output layer.The experimental results show that: 1.Attention mechanism can effectively learn important context-dependent relationships in the sequence,it performs better than the classical deep network LSTMs in Chinese sentiment classification,therefore,it is feasible to widely use the attention model in NLP tasks,which has a simple structure and easy training.2.Compared with the standard LSTM network,the Bi-LSTM network can better retain and control the context information of the sequence,thus it is more reliable.3.Compared with single models,the hybrid model performs slightly better in Chinese sentiment classification tasks,which shows that the hybrid model can overcome the defects of the single model in some aspects and it can improve the performance of the classifier.The hybrid model proposed in this paper is a generic framework that can be used in other similar NLP tasks.