Study On Credibility Of Online Products Reviews

Online product reviews refer to the consumers’opinions about the products, sellers and service, which is published on the e-commerce websites, forums, microblog or other commentary websites. The reviews gather a large number of consumers’ opinions, and at the same time, it has gradually became an important basis for consumers to make purchasing decisions and significantly affected the sales of the products. Ensure the credibility of reviews is crucial to consumers, sellers and the market stability. In view of this, this paper stands on the credibility of online product reviews and aims at filtering the lower credible reviews by machine learning to offer valuable reviews for consumers to make the purchasing decision.Based on the deep analysis of the online product reviews’ characteristics, also with some related works, this paper made an empirical analysis on the credibility factors by questionnaires. According to the results of the empirical analysis, this paper selected content integrity, sentimental balance, review timeliness and credit rating of the publisher as four features. This paper used the CRFs model to identify the content integrity and sentimental balance feature. The aspect of content integrity selected word, speech, Hownet sentiment words and ontology features to identify the evaluation object. The aspect of the sentimental balance selected word, speech and Hownet sentiment words to identify the evaluation words and calculated the sentiment value as the sentimental balance feature. The method is applied in opinion elements evaluation task of the Chinese blog in NLP&CC2013. The result of loose evaluation is ranked the first among21institution. The result of strict evaluation is ranked the third. It proved the effectiveness of the method.In the credibility classification experiments of product reviews, it used CRFs as reviews credibility’s classification model, and conducted feature combination experiments to get the best feature combination. The experiments achieved significant results, and the correct rates of the classification model are all above75%. The research results of this paper can improve the existing artificial effectiveness evaluation method, thus offer new methods and thoughts for optimized filtering of the online reviews.