Topic-Field-Specific User Influence Research And Implementation In Microblogging Platform

Microblogging,as one of the most popular social network platforms in recent years,is charactered by its number of users,user activity,and the vast speed of information generation and propagation.Microblogging is different from other traditional social network platform by its social media feature.It allows everyone on the network to create and spread their opinions in a convenient and fast way.Among all the users in Microblogging network,few influential users act as very important roles in the propagation of information.They can influence public opinion and create topics that can draw a lot of attentions.Measuring users’influence can find the potential values hidden in the vast and mess information in the complex microblogging network.Finding influential users provide an effective solution in fields such as advertisement,marketing,recommendation and government supervision.Recent researchs have find different ways to measure the user influence. Statistics such as number of fans, number of tweets, retweet numbers are often considered as important factors of user influence. Web Page Ranking algorithms are considered more accurate to measure user influence. They use random walk in the social network to simulate the reading behaviors in microblogging network. The current research considered more about the overall influence of users, rather than divide the user influence by different topics or fields. But user influence varies in different topic fields. In this paper, we present a Topic Field Specific microblogging user influence algorithm. We analysis the users text content he or she published to calculate the user’s topical feature, thus get the topical value of each user, and the topical similarity between two users. We build the retweet network to run the Topic Field Specific User Rank, thus get the user influence rank. The main work and innovations are as follows:First of all, we study the information propagation on the microblog network to find its chacteristics and rules. We studied the user interactions on the network to find out how people influence each other and how opinions spread on the mircoblogging network.Secondly, we promote our algorithm architecture. There are two main parts of the algorithm. Firstly we aggregate all the text type content each user published, and calculate the topical feature vector, and then get the user topical value, and topical similarity between users. Secondly, we build the retweet network of the users and run the Topic Field Specific User Rank, which is a variation of Page Rank algorithm, to get the final topical user rank. We use the topical similarity we gained in the first part to modify the transition probabilities and the topical value to modify the random jump in the network.Finally, we evaluate our algorithm in a Hadoop distributed platform. We choose Sina Weibo, the most popular microblog platform in chinese mainland. Firstly we use the Sina API to mine the users profile and weibo data as resources. Secondly we choose a specific topic field and gathered key words to build the key word set so as to calculate to topical influence. We programed in Map Reduce method to deal with large scale of data.