Online Reviews Sentiment Analysis Research Based On Combinations Of Methods

In recent years,with the rapid development of Internet technology and extensive application,e-commerce websites,micro blogs,news websites,etc.have gradually become an indispensable part of people's lives.These websites usually have a large number of online reviews,which contain great value and have become an important source of information for consumers and enterprises.Due to the large amount of information and unstructured online reviews,text mining technology is intensively applied.Sentiment analysis is a text mining method to analyze the sentimental views of online reviews.It mainly carries out two tasks: sentiment polarity classification,analyzing the sentiment polarity of text sentimental views,including positive,negative and neutral;aspect identification,identifying the specific aspects of sentimental views,which can be any attribute or feature of a specific entity.However,most traditional sentiment analysis methods focus too much on the sentimental words in the text and the nouns associated with them,and seldom consider the influence of the other parts of the sentence.Based on this viewpoint,this dissertation attempts to find a practical way to improve the performance of sentiment analysis.In this dissertation,a combination of some traditional methods is proposed which focuses not only on obvious sentimental words and nouns,but also on the other parts of the sentence.For both the sentiment polarity classification and the aspect identification tasks,the dissertation verifies that the proposed combination method is effective in improving performance.The research is conducted at two different levels.First of all,in the sentence level sentiment analysis,based on the characteristics of the sentiment analysis,this dissertation distinguishes between explicit and implicit sentiment opinions.In addition,considering the advantages and disadvantages of existing sentiment analysis methods,a combination method based on dictionary and machine learning is proposed.The core idea is to combine the advantages of lexicon method in the classification precision rate of explicit sentiment and machine learning method in the classification recall rate.Next,based on the construction of sentence level combination method,it is introduced into the more fine-grained aspect level,and a candidate opinion extractor based on dependency syntax rules is established.In addition to extracting specific adjective phrases and noun phrases,the part of the sentence without sentimental words is considered as another type of candidate opinion,and then a supervised machine learning discriminant model is cascaded to obtain aspect categories.In the experiment of sentence level,the overall performance of the proposed combined sentiment polarity classification method is better compared with the traditional baseline methods;in the experiment of aspect level,the proposed method also performs better than other existing methods on the aspect identification task.The experimental results also verify the effectiveness of the proposed sentiment analysis method.