Sentiment Analysis Of Automobile Product Reviews Based On Fine-grained Opinion Pairs

With the development of social networks,e-commerce and mobile Internet,billions of people actively participate in the development of Internet,making the amount of Internet information increase with each passing day.Some users are happy to share their views by e-commerce sites,micro-blog or forums,making the comment information overwhelming.This type of information reflects the people's different attitudes and different sentiment tendencies,such as affirmation,approval,criticism,questioning and so on.Through the analysis of product reviews,not only allow ordinary users to better understand the market evaluation of certain products to make more accurate decisions,but also allow businesses use such results to obtain the most intuitive market feedback to make more targeted decisions.There is no doubt that figure out the clues only by manual method in the massive comment data is not easy,so it is urgent to use computers to quickly and accurately mine the valuable information in these review data.Reviews on the Auto Products Forum website are expressed in various ways and forms,and some do not even have any valuable information.For example,"I would like to buy second-hand classic Focus,who would like to sell? Please reply!" Another example is "I have concerned about the EXCELLE for a period of time,I was very satisfied on the whole and also went to the 4S shop to see the real car,but there is no discount at present.Recently in urgent need of the car,I want to know where there are concessions in Chengdu,thank you heroes!" Only part of the review has sentiment,and if it can be extracted out through analysis,it will greatly reduce the cost of information acquisition.Confronted with the sentiment analysis problem of the automobile product reviews,in this paper,we provide a sentiment analysis method that is based on fine-grained opinion pairs,which can automatically extract the opinion pairs in the comment texts,analyze their sentiment tendencies and carry out the results visually.First of all,take a web crawler approach to crawl the comment data on the relevant Auto Products Forum website and preprocess the data.Then use LDA topic model to extract the subjects contained in the text,and then summarize,categorize and construct domain-specific feature word dictionaries,and improve the sentiment dictionaries and the adverbs dictionaries.By analyzing the possible collocations of the various parts of speech in the automotive field,this paper designed and perfected the extraction model of various opinion pairs and proposed the calculation method of sentiment polarity with the different opinion pairs.Based on the extracted elements,a vector space model of the weighted text is constructed to represent the product reviews.And finally,we use the KNN algorithm to distinguish the sentiment tendencies of the test set.The experimental results show the accuracy rate of the method provided in this paper is up to 94%,which has certain application value.At the same time,this article presents a visual display of fine-grained opinion pairs in the automotive product reviews and the sentiment propensity of the product reviews,making the results of sentiment analysis more intuitive.