Research On The Method Of Extracting Opinions Based On Product Reviews

With the mellow of Internet technology, all kinds of web-based electronic platforms obtain great development and popularization. People can express their views and opinions through the network media, which contains large amount of useful information. Product reviews is an important part of user-generated text information, how to mining information from it fast and accurate has become one of the hot issues in Natural Language Processing related field.In this paper, we study the method of extracting chinese opinion from the product reviews under the machine learning framework with various features and knowledge. We focus on the issues of opinion element recognition,opinion relations extraction and dynamic polarity classification. Concretely, this paper mainly studies from the following three aspects:(1) Opinion elements recognition under conditional random field with knowledge base : Opinion elements are divided into explicit opinion elements and implicit opinion elements, the recognition of implicit elements is always considered as the bottleneck of the recognition of the opinion elements. We solved it under conditional random field framework with morphology, part of speech, the position information and context information as features,the experimental results were satisfactory. In order to identify implicit opinion elements, we use statistical methods construct collocation repository, then in conditional random field framework determined position of implicit opinion elements. Finally, Use the collocation repository to determine attribute implicit. Experimental results show that multiple features and knowledge are benefit for improving the performance of opinion elements recognition, especially the recognition of implicit opinion elements.(2) Opinion relations extraction under support vector machine framework : opinion relations can be divided into Aspect-of relations which consist of product brand and product aspects and Aspect-Evaluation which consist of product aspects and evaluation. The extraction of complex opinion relations is always considered as the bottleneck problem of opinion relation extraction. This paper argues that the complex opinion relation is composed of many simple opinion relations. In order to extract the relations of opinion, this paper is to deal with the problem in the form of a classification problem. Firstly,construct simple relation candidates, then under the support vector machine(SVM) framework with part of speech, dependency relations and distance information as features construct opinion relations extraction system. The experimental results show the effectiveness of the proposed method.(3) Sentiment polarity classification based on dynamic word’s association words lexicon. We take the product aspects in product review sentences as the main body,the sentence granularity of emotional polarity classification problem is transformed into the word size of sentiment polarity classification problem. In particular, the emotional polarity classification of the dynamic polarity words is the bottleneck problem for sentiment polarity classification. To solve the problem, this paper presents a dynamic polarity based association Dictionary of sentiment polarity classification method, The experimental results show the effectiveness of the proposed method.The experimental results show the effectiveness of the proposed method.