Research On Deceptive Reviews Detection And Recommendation Algorithm Based On Web Mining

Demand is the driving force for technological progress in all sectors of the industry,as is the Internet industry.With the increasing demand for the Internet,Internet technology has been further developed and applied.While the progress of technology to make people's needs to be met,it brings people's troubles.Although convenient network access makes people's life easier,but also brings people some problems,such as information overload,cyber fraud,and so on.In the face of the explosive growth of information,people have put forward new demands on the Internet technology.Thus,in order to solve these problems,such as information overload and deceptive information,Web application technologies,such as Web data mining technology,the information technology retrieval whose representative is search engine and as the information filtering technology whose representative is the recommendation system and so on,have been gradually explored and obtain rapid development.In order to improve the real reliability of network data and provide more personalized Web services,this thesis will study the false comment recognition technology and recommended technology,has mainly done the following aspects of the work.Firstly,a method based on semi-supervised learning is used to identify fake reviews.Based on the construction of a small number of annotation sets,features are defined,and genetic algorithm is used for feature selection,the improved co-training technique is used to make use of the unlabeled data,and recognize fake reviews according to the two dimension of characteristics of the comment feature and the commenter's characteristics.Experimental results show that the deceptive reviews detection method proposed in this thesis has a good performance in the recognition effect.Secondly,a collaborative filtering recommendation technique based on project attribute preference mining is designed.A new probabilistic dilution method is put to use to mining the multi-tagged item attribute.In the process of user similarity calculation,the doublethreshold similarity calculation method is adopted,and the project attribute information is put into the users' preference mining process,to a certain extent,ease the sparse data problems.According to the experimental results,collaborative filtering recommendation technology based on project attribute preference mining is superior to traditional recommendation technology in recommendation effect.Thirdly,a personalized hybrid recommendation approach is designed.This method uses more comprehensive data information to construct more detailed user and product model,and through similarity calculation a more similar and diverse neighbor set has be got with the score data.The design negotiation model strategy integrates the recommended list of recommendation methods,and improve the credibility,accuracy and diversity of the recommended results.The experimental results demonstrate the feasibility of the personalized hybrid recommendation method.