Research Of Sentiment Analysis Oriented To Product Reviews

With the rapid development of online social media and e-commerce,a large number of product review information has appeared.The user feedback hides in these product reviews.Mining the valuable information,especially analysis the sentiment polarity of product review information,which can provide reference for users to purchase,and at the same time help merchants improve product quality and improve operating strategy for business.As a branch of text mining,sentiment analysis technology has become a research hotspot gradually.Based on the film review data,the paper conducts coarse-grained sentiment analysis and fine-grained sentiment analysis of product reviews.Combining LSTM and CNN,a coarse-grained sentiment analysis model is constructed.On fine-grained sentiment analysis,the paper focuses on product aspects extraction and clustering methods.Finally,a sentiment analysis prototype system of film reviews is designed.The main research work of this paper is as follows:(1)Data collection and preprocessing.Collecting reviews of multiple movies on the Douban website with web crawler technology.After collecting reviews,the word segmentation tool is used to perform word segmentation and part-of-speech tagging and the stopword dictionary is used to remove the stop words.Then the word vector representation technology is used to realize the vector representation of the text,and the data preprocessing is completed.(2)Coarse-grained sentiment analysis.Combined the word vector feature extraction method with the SVM classification model,the traditional machine learning method is used to classify the emotions.In order to improve the accuracy of classification,based on the research of LSTM and CNN deep learning methods,the polarity classification model of LSTM_CNN is proposed,which combining the two methods and verified by experiments.(3)Fine-grained sentiment analysis.The research focuses on the extraction and clustering of aspect words.Firstly,the LDA topic model is used to model the topic of the review text,and the aspects are extracted based on the frequency and the evaluation relationship.Then clustering the aspects by calculating semantic similarity using the cosine similarity with the the aspects seeds obtained by LDA model,and build topic-aspects dictionary.After text aspect labeling is completed with the topic-aspects dictionary,the LSTM_CNN classification model is used to realize the aspect emotional polarity statistics.(4)Film review sentiment analysis prototype system design.Based on the film review data obtained in the paper,the film review sentiment analysis prototype system is designed to realize the function of coarse-grained sentiment analysis and fine-grained sentiment analysis on short comment texts,so as to visually display the polarity statistical results of users' opinion sentences and various aspects in the text.