User Interest Profiling And Information Recommendation

With the rapid development of Internet technology,information overload has become a serious problem.Both producers and consumers of information are faced with enormous challenges.Personalized information recommendation service is an important measure to improve the accuracy of information acquisition.Personalized information recommendation service is performed by accurately profiling the interest of users.However,users' interest is unstable and broad.The method of interest profiling and the modeling techniques are restricted by the form of data.The users' interest profiling is the important and difficult part of the whole personalized recommender system.Micro-blog,the representative of social applications,is increasingly changing the way of our life.The enormous amount of user data in it has important value.The texts published on it reflect the users' preferences so that users' interest profiling based on micro-blog is feasible.Based on this background,we research on the active users' interest profiling and the information recommendation based on the users' interest model.Firstly,we complete the identification and detection of active users using the SVM classifier.Secondly,we extract users' interest tags using the text weight calculation method based on AHP and establish the tag-based vector space model.We profile users' interest using LDA topic model and sentence vector respectively and propose a text similarity calculation method based on the integration of these two models.Thirdly,we improve collaborative filtering recommendation algorithm using spectral clustering and improve the recommendation performance.Finally,we apply the results of this study to news recommendation system.Through the implementation of the system,we prove the feasibility and effectiveness of the algorithm.The innovations of our study are:(1)We proposed a text weight calculation method based on AHP taking advantage of the characteristics of micro-blog text.(2)We proposed a text similarity calculation method based on the integration of LDA and sentence vector model.(3)We improve the performance of collaborative filtering using spectral clustering.