Discovery Of Social Network User Similarity Based On Bayesian Network

User similarities in social network as an important research field can be used in product recommendation and user relationships evolution. User similarities not only depend on the topological structure of user behaviorial interactionsbut also the dependence degrees between users. In the paper, we adopt Bayesian network (BN), an important and popular probabilistic graphical model, as the framework to discover user similarity. However, with the rapid development of social networks, the social network data show the characteristics of massive, distributed storage and dynamic change. The traditional methods of BN construction and referring can’t meet the application requirements. Fortunately, the appearance of MapReduce programming model offers excellent technical support for the analysis and management of massive data, which makes it possible for discovering user similarities from massive social network data.To represent the direct similarity relationships between social network users, we propose a BN of social network users, called social user BN and abbreviated as SUBN. The construction of SUBN mainly includes two parts, namely, DAG construction and CPT learning. In DAG construction, through map and reduce process, we obtain the edges of SUBN by reading datasets in parallel and then we can determine the direction of edges. Finally, we obtain the DAG In CPT learning, we obtain CPT of SUBN by computing conditional probabilities in parallel through map and reduce process. To support the high efficient SUBN inference and user similarity algorithm, we give an HBase based distributed storage and SUBN parallel inference method.To find the indirect similarity relationships between social network users, we give a novel SUBN-based user similarity degree combing the influence of SUBN structure an SUBN inferences. In order to accelerate the SUBN inferences, we propose a MapReduce-based SUBN inference algorithm. In the end, according to our SUBN-based user similarity we find the user similarity in social networks.We tested our approaches on Hadoop platform. The result of experiment indicates the correctness and execution efficiency of our method.