Research And Implementation Of Fine-grained Sentiment Analysis Based On Commodity Reviews

With the rapid development of Internet technology,commodity reviews on ecommerce websites are also growing exponentially.How to mine valuable business information from the mass of comments has become one of the most difficult problems at present.Therefore,text sentiment analysis technology came into being.As one of the main research contents of text sentiment analysis,fine-grained sentiment analysis is used to analyze text at the attribute level.According to the results of fine-grained sentiment analysis,it can provide the basis for businesses to improve goods and provide reference for consumers to choose.In the study of fine-grained sentiment analysis with commodity reviews,aspect extraction and sentiment analysis are the most important tasks.In this thesis,we focus on the problems including a great deal of manual labeling effort,low accuracy in unsupervised learning approaches for aspect extraction,poor adaptability,low coverage in the field of existing sentiment lexicon in sentiment analysis.The work of this thesis is as follows:1.Aiming at the problems of a great deal of manual labeling effort,low accuracy in unsupervised learning approaches for aspect extraction,a new aspect extraction approach based on affinity propagation clustering algorithm and point mutual information pruning is proposed.Firstly,this approach extracts candidate aspects according to a pre-defined set of syntactic rules.Secondly,using word embedding stands for candidate aspects and executes affinity propagation clustering algorithm.Thirdly,we use the similarity between cluster center and the indicator to prune non-aspect clusters and point mutual information to prune aspect clusters.Finally,we find out the aspects.Experimental results show that the approach can effectively improve the accuracy of aspect extraction.2.Aiming at the problems of poor adaptability,low coverage in the field of existing sentiment lexicon in sentiment analysis,a new fine-grained sentiment analysis approach based on sentiment lexicon.Firstly,this approach uses TF-IDF algorithm to train the words in the comments to obtain candidate seed word set.Secondly,candidate seed word set are intersected with the basic sentiment lexicon and the ones with high TF-IDF are selected as the seed words.Thridly,we use SO-PMI algorithm to identify sentiment words and combine with the basic sentiment lexicon and Boson NLP sentiment lexicon to obtain sentiment lexicon.Finally,we use aggregate sentiment score by combining with aspect word set to judge the sentiment orientation of aspects.Experimental results show that the approach can effectively improve the accuracy of fine-grained sentiment analysis.3.We designe and implemente a fine-grained sentiment analysis prototype system.The system can automatically extract aspects from commodity reviews and visually display sentiment orientation of the aspects.