Study On Recommendation Model And Algorithm Based On Uncertain Statistics And Uncertain Set

As the electronic commerce gradually into the era of social business,there are a wide range of methods for consumers to share and obtain information,meanwhile,these convenient functions also bring about a problem of information overload.Recommender systems provide consumers with an effective technique to solve the problem of information overload.However,it is usually uneasy to obtain sufficient information for recommender systems from the real e-commerce sites,and characterize consumers' preference based on the ratings.Online reviews are feedback by consumers after buying the products.Therefore,it is more accurate to mine the consumers' preference from online reviews,and also an effective way to solve the problem of spairty.But there are undetermined problems to describe preferences from reviews because of unstructured and free style of online reviews.However,the characteristics of online reviews are unstructured,free which increase the uncertainty of describing items.In order to solve this uncertain problem,the individuals and group profiles were designed based on the objects of recommender systems by uncertainty theory,the techniques and methods of data mining,using online reviews and ratings as data source.We also carried out experiments for further research.The details are as follows:Firstly,according to the problem of users' uncertain preferences in reviews,the sentiment analysis model was designed based on the uncertainty theory,and then it was applied to individual recommender algorithm.The sentiment analysis model for users' was built based on the sentiment strength and sentiment polarities that were characterized by uncertain variable and uncertain set,respectively.According to the sentiment analysis model,we proposed a new similarity calculation formula,and applied it to individual recommender algorithm.Secondly,for the problem of multiple criteria sentiment analysis in reviews,we explored multi-criteria sentiment analysis model based on uncertainty theory and userinterest.Then the model was applied to individual recommender algorithm.The total satisfaction of users' for an item was introduced to compute the number of multi-criteria by the method of linear regression.We built the multi-criterion domain emotion dictionary to keep balance of the number of positive and negative emotions words in data sets.The multi-criteria sentiment analysis model based on uncertainty theory and user-interest was applied to design the individual recommender algorithm.Lastly,in order to provide group with recommendation service,we explored group recommendation model by integrating ratings.The group recommendation model was designed to solve the preferences conflict for group users' by the methods of collaborative filtering and uncertain statistics.And then the group recommender results were obtained by the proposed group model.Experiments show that our approaches achieve a significant improvement of improving the accuracy and alleviating the data sparsity problem,respectively.The data source of experiments was formed two parts.One was crawled from one region of HankowThames,including Sport Fitness,Hotel,and Restaurant.The other was MovieLens and Restaurant datasets which were provided by GroupLens and Yelp,respectively.