Research On Bayesian Network Based Recommendation Techniques In Online Social Networks

When confronted with massive data, users need to use recommendation technologies to find the contents they are interested in quickly and accurately. However, the recommendation technologies in online social networks may suffer the issues of cold start and data sparsity which often make the models not match the data and decrease the precision of recommendation. In this thesis, the recommendation technologies in online social networks are studied, and the recommendation methods based on bayesian network which can improve the recommendation quality are designed to alleviate the problems of cold start and data sparsity. The recommendation methods can also solve the problem of low coverage and protect users' prvacy when users can not provide sufficient feedback.Effects of some factors such as social relations, historical ratings and feedback on recommendation methods are especially studied in this thesis. The recommendation method based on the explicit and implicit feedback in online social networks is put forward. Bayes theorem is applied to recommendation system and the technology of probabilistic matrix factorization with explicit feedback, implicit feedback and trust relationship information is used to decompose the trust relationship matrix and the rating matrix. The parameters of the training model are optimized, and the users are provided with ratings. The experimental results show that the recommendation method in online social networks can obtain user preferences effectively and provide a large number of highly accurate ratings. The recommendation in online social networks has a good performance.In addition, a trust-driven recommendation method by constructing Bayesian network in online social networks is proposed in this thesis. The method can be used to predict users' ratings about the products and provide the users with appropriate products accurately by calculating the prior probability distribution of users' ratings and the joint conditional probability between the friends. The experimental results show that the trust-driven recommendation method in online social networks is more effective than the traditional recommendation methods in rates of coverage and precision. The method can solve the problem that the accuracy of the recommendation is not high to a certain extent.