Researches On Recommendation System Based On Social Network Services

With the development of Web2.0technology, social network has become an important channel for people exchange and access to information. This means that the amounts of users’informational data and user-generated data are growing fast. At the same time, the development of mobile terminal technology allows users to upload textual, pictorial, musical data and other multimedia information anytime, anywhere. In order to solve the problem of mass data, it is meaningful that we researches on how to recommend to a specific user according to its tastes, how to help user receive his/her photos/items he/she will potentially interested in, and also maximizing the value of information which is not so wide spreading.We need to solve mainly three problems when building a recommendation system:1) how to define the recommendation problem;2) how to obtain the similarity between users;3) how the strategies could be designed when recommending to users. These three problems interact with each other, so the design of recommendation system’s framework is extremely complex. Committed to addressing the above three problems, and constructing a reasonable recommendation system framework, the main research works in this paper are as follow:(1) We regard the recommended problems as learning problems, so the kernel of recommending problem is to obtain users’interest model as accurately as possible, the problems we have solved are:1) User interest modeling. We think the user likes the item only for two reasons:a) he/she is interested in the content of the item; b) he/she is interested in the item for his/her friends are interested in it.2) The problem of mapping between social tag and topic. The social tag is freely used by users and has lots of form. We utilize WordNet to calculate the semantic relationship between social tag and topic;3) Combining recommending with keyword searching. The items firstly returned by keyword searching, and then it can improve users’satisfaction by sorting the items according to the user’s preference information.(2) We regard the recommendation problems as prediction problems to learn user’behavioral histories. In daily life, users’interest is changing over time, the change of users’behavior can reflect change of users’interest in some degree. So we can "observe" clues for users’interest changing from his/her behavior. Take check-in service in the social mobile network for example, we can only observe the change of users’ check-in behavior, when they checking in and which location they check in at. While the reasons for checking in is various, such as the happening of social events, the influence by their friends, or the good or bad mood caused by delicious food and trouble things respectively. Based on the above analysis, we introduce the Hidden Markov model to building a probabilistic mapping between the unknown reasons and the observable behaviors.Based on the above researches, we respectively use the C#language and Matlab tools to design and implementation of two social network recommendation system based on Bayesian model and the HMM. After our evaluation on4social networking data sets, the results show that proposed systems can solve the recommending problems in some degree and also improve the user satisfaction.