Study On Opinion Mining Of Product Reviews From Online Shopping Platforms

With the rapid development of the Internet and E-commerce,product reviews have captured more and more interests from customers.However,customers are nearly impossible to extract useful information for purchasing decisions from large quantities of product reviews manually.To solve this problem,this paper focuses on opinion mining,the task that extracts opinions of customers on a particular product from product reviews and then organizes all opinions as a product opinion summarization efficiently.A product opinion summarization of a particular product gives several types of information: useful information of purchasing decision making for customers,meaningful information of customer preferences for merchants,information on how to improve the product for manufacturers and information of product quality for managers for online shopping platforms.To perform opinion mining,this paper makes lots of effects including summarization of related work,discussions of related text mining techniques,methods for generation of product opinion summarizations and experiments designed to validate effectiveness of the proposed framework for generation of product opinion summarizations.Moreover,various product opinion summarizations proposed in related work are also discussed.According to related work,the tasks of generation of product opinion summarization include recognition of product features,recognition of sentiments and process of matching product features with sentiments.Text representation,feature selection and sentiment analysis are discussed in this paper.Moreover,a new feature selection method termed Marginal Utility Score(MUS)is proposed to recognize people features(kinds of terms representing people)in product reviews.This paper describes a detailed framework for generation of proposed product opinion summarization.The method covers processes from input of product reviews to output of final product opinion summarization.Contributions in this paper are described as follows.Firstly,a new feature selection method is proposed to recognize people features in product reviews to understand whom the products are bought for so as to make more precise recommendations.Secondly,co-occurrence analysis is applied to discover co-occurrence relationships between product features and people features to understand customer preferences in order to make more precise reommendations.Thirdly,the proposed product opinion summarization offers more information including total sentiment scores for product features and people features(used to sort features),sentiment intensities of corresponding sentiments(used to sort sentiments)and co-occurrences(used to discover customer preferences).Moreover,the proposed product opinion summarization is an interactive HTML document.Users are allowed to select preferenced information by pressing the buttons in the document.A real product review dataset is generated to test the proposed framework.Experimental results indicate that the proposed framework is capable of generating useful product opinion summarization which contains rich information and is easy for users to read.