Temporal-based Ranking In Heterogeneous Networks

Ranking is a fundamental task for network analysis, benefiting to filter and finding valuable information, and become one of the hot topics that the Internet and academics concerned. The conventional rankings focused on the single factors of homogeneous network, namely, all nodes type in the homogeneous network are same, thus the ranking possess the strong determinacy of the factors. For the heterogeneous networks, that composed of multiple types of node and complex reliance structures, there is a mutually information imbalance relationship between different types of nodes. Hence we apply the information flow propagation to optimize the ranking results. Meanwhile, the varation and nondeterminacy of the time factor can affect the optimization of ranking results as well, and can introduce error or bias into extracting and mining the valueable information.In this paper, we proposed a hierarchical ranking model on heterogeneous network based on the time factors' varation and the ranking's nondeterminacy of homogeenous nework. Ranking on weibo for instance, we make full use of the diverse ranking in homogeneous network to initiate the web pages, weibo and users ranking results, thereby obtaining the pre-ranking results of the heterogeneous network. According to the time-based propagation of weibo, we use the logistic regression method to fit the weibo's life cycle curve, and gaining the varition of the time characteristics. After that we optmize the weibo initial rank by means of the weibo temporal weight. Finally, combining the ranking results of web, web and user to balance the information between heterogeneous network based on the infomation flow between different type of the node, and to obtain the optimal ranking results consequently. Hence, finding the valuable and popular weibo information.The experimental tests take sina weibo for instance, and we crawl 1.7 million users' original microblog data through web crawlers. According to the experimental results, we can prove that the TemporalHeteRank method can effectively improve the ranking's accuracy, real-time and reliability. Then we apply the TemporalHeteRank to hotspots detecting application, and compare with the existing hotspots detection model, thus to analyze and prove the feasibility of proposed weibo ranking approach.