Research On False Analysis Of Cross - Domain Product Reviews Based On Genetic Algorithm

With the maturing of the electricity on the Internet, many people will choose to shop online.At the same time the product reviews will influence the decision people buy goods, which has led businesses to deliberately write deceptive reviews to improve the sales of goods or undercut rivals. Then, it is necessary to identify deceptive reviews and the study on identifying deceptive reviews becomes a hot issue of text sentiment analysis. However, the current methods have the low recognition accuracy and the high complexity. And the study on identifying deceptive reviews is difficult when the lack of labeled data or less. Thus, this thesis studies the problem of identifying cross-domain deceptive reviews based on transfer learning, genetic algorithm and mapping techniques.Firstly, aiming at the problem of identifying cross-domain deceptive reviews, this paper selects the optimal feature set from known source deceptive reviews based on genetic algorithm. This thesis processes the reviews according to deceptive feature of deceptive reviews. Then, the thesis encodes the structured reviews into chromosomes. After designing the fitness function based on logistic regression, this thesis selects the optimal features according to the process of genetic algorithm. The optimal feature set provides a support for the reduction of the complexity of the identifying deceptive reviews. Moreover, differences between deceptive reviews and real reviews are presented by an experiment.Secondly, based on the optimal feature set, in order to identify deceptive opinion in unknown domain, we propose a method of detecting deceptive opinion on transfer learning. This method constructs the feature incidence matrix between known and unknown domain by the similarity of documents. According to the mapping function, sentiment classifiers are trained. Then, experimental results demonstrate empirically the feasibility and advantages of the method proposed.Thirdly, this thesis designs and implements a prototype system for the identification of deceptive reviews, which provides a platform and foundation for further research on the identification of deceptive opinion.