Article Preview
TopIntroduction
With the rapid development of social networks, the scale has increased sharply, and the basic data information of users has increased dramatically in social networks, social networks is a data source with massive data. Through the analysis and research of massive social network data, the potential value in the data can be discovered, which can help people study the interaction behavior between users and the common behavior of similar users. Discovering social network user similarity helps to understand the evolution of relationships between users, and at the same time, a method is refined to find the best product recommendation (Han X, et al., 2016; Wang W, et al., 2015; Li J, et al., 2019;Ma T M, et al., 2015;Yan G, et al., 2019;Luo E T, et al., 2019;Zhang S S, et al., 2019).
User similarity in social networks, as an important research field in social media data analysis, can be well applied to research on product recommendation and user relationship evolution in social networks. The similarity between users not only depends on the network topology, but also closely related to the degree of dependence between users. However, with the rapid development of social networks, social network data presents the characteristics of massive and distributed, which makes traditional Bayesian network construction and reasoning methods unable to meet application needs. Fortunately, the MapReduce programming framework provides good technical support for massive data analysis and processing, making it possible to discover user similarity in massive social network data. In order to discover the direct similarity between users in massive social network data, this paper proposes a MapReduce-based social user Bayesian network (SUBN) construction approach. The construction of SUBN mainly includes Directed Acyclic Graph (DAG) construction and Conditional Probability Table (CPT) calculation. In the construction of the directed acyclic graph, the data set are read in parallel through the map and reduce functions, which determine whether there are edges between user nodes and determine the direction of each edge according to the corresponding metric formula, and the directed acyclic graph is finally obtained. In the learning part of the conditional probability parameter table, the data set is read in parallel through the map and reduce functions, and the conditional probability parameter table corresponding to each node is calculated. In order to better support the efficiency of Bayesian network reasoning for social network users and user similarity discovery algorithms, a Bayesian network distributed storage and parallel reasoning method based on HBase is proposed. In order to discover the indirect similarity between users in massive social network data, a MapReduce-based user similarity measurement method is proposed by combining the topology structure of SUBN and the probabilistic reasoning of SUBN. Finally, according to the BN user similarity measurement method, the user similarity in social network data is found.