| Summary: | 碩士 === 東吳大學 === 資訊管理學系 === 104 === With the change of cross-people communication, people use online social network widely to make social connection closely with friends who are online or in real world. Many famous brands adopt Facebook Fan Page as main interactive marketing media with their users. Online social network media are getting more convenient and important. Due to user’s sparse interactive behaviors on online social network, the effect of social recommendation is not good. On the other hand, the new user/item without any historical record causes the insurmountable cold-start problem. Users are spending much time to deal with huge and complex data at the same time.
This research studies on movie recommendation system by using collaborative social network filtering in order to effectively reduce the searching process and increase benefit of filtering mass notifications from social media. By analyzing varies social interactive behaviors of users and adopting Collaborative Filtering Based Social Recommender Systems, it will find the nearest neighbor of active users through integrated user similarity. According to the movie selected by friends with similar social habit, we provide users with the customized movie recommendation model which meets their interest. For new members or one-time members, the popular and influential endorsers provide them with movie recommendations to increase the chance of reading and interactive with social group.
We estimate the method generalized from the data of Facebook Movie Fan Pages. The experiment result shows that the Integrated User Similarity Method is more precise than the current Collaborative Filtering Based Social Network Recommendation in terms of Precision rate, Recall rate and F1-Measure. The method dramatically increases the performance of recommendation in social network. Through the preference of friends with the social similarity habits, it will overcome the situation of sparse data. Through the popular and influential endorsers, it will effectively solve the problem of Cold Start.
|
|---|
