Font Size: a A A

Recommendation Research Based On Social Activities And Friend Networks In Social Networing Sites

Posted on:2015-03-07Degree:MasterType:Thesis
Country:ChinaCandidate:Y Y DengFull Text:PDF
GTID:2308330464455518Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
The global number of Internet users has risen sharply with the rapid development of the Internet. In the Internet world, people not only acquire information but also provide information to others. As a result, how to dig useful information for users to avoid the bad experiences brought about by information overload is in a spotlight in both academia and industry. Personalized recommender system has emerged to solve such problems. It is designed to help users make decisions quickly under large amount of information, or recommend content of interest to users by analyzing and mining user interests. These features of recommender system can enhance the user experience. In recent years, most of the Social Networking Sites(SNSs) have implemented the prototypes of recommender systems, which can recommend friends or recommend content of interest to users. With the expansion of the scale of users in SNSs, us-er-generated content is in sharp increase and user’s friend network is expanding. Us-ers in SNSs are troubled with two problems:(1) missing topics of interest due to in-formation overload; (2) finding it difficult to filter out the target friend to share topics with due to the large number of friends.Collaborative Filtering(CF) technology is one of the most successful personalized recommendation techniques. The traditional CF recommendation method has faced some limitations when applied to SNSs as a result of the unique features of SNSs. In recent years, more and more studies focus on social recommendation, a large majority of which are about how to incorporate social contextual information into CF model to improve recommendation quality. This dissertation starts from the above two prob-lems, focusing on generating recommendations by connecting users with common interests and helping sharers to filter out the target friend to share content with. The dissertation makes the following contributions:● SoSAN—a recommendation algorithm based on user’s social activities and friend networks, which combines the attention weight and similarity among user pairs to build influence weight from each user to other users. The similarity method in SoSAN adopts the improved Jaccard method that expands the weight of com-monly commented behavior by a user pair. Experimental results in a real Social Network show that the recommendation quality can be improved when recom-mending based on influence weight and the improved similarity method perform better than the standard Jaccard.● ComL—a linear model, which is used to recommend a friend list for sharers to help them filter out target friend when sharing a topic. It is the combination of recommendation based on SoSAN algorithm and on sharer’s share histories. Ex-perimental results in a real Social Network show that ComL can perform better hit ratio.● AOPUT—a general recommendation framework designed to use in typical SNSs. It contains two core functions:(a) recommending a specific topic to users who may have an interest in this topic based on SoSAN algorithm, (b) recommending a friend list to sharers when they want to share topics with friends based on ComL model. The dissertation describes the functional components, detailed workflow and data model design in detail, and also analyzes the generality and responsiveness of the framework.
Keywords/Search Tags:Social Recommendition, Social Network, Collaborative Filtering, So- cial Activities, Friend Network
PDF Full Text Request
Related items