Research On Topic Model Based Personalized Expertise Ranking Algorithm In Community Based Questioning Answering

With the popularity of online social network, more and more web users are starting to publish information, to propagate resource and to maintain social relationship with different people through online social network widely. Due to the diversity of participant community and content format, the main producer of web resource and information has been transferring from the traditional website to the ubiquitous social users. An amount of social users oriented documents, images, sound and videos are constantly and increasingly emerging on the web, which composed the significant treasure in the era of big data. Hence, the issue that how to retrieval the users’target content or resource is significant.Search engine has been the main source for discovering user generated content with authority, not only on content-centric multimedia networks but also on human-centric online social networks. Many studies have demonstrated the power of graph based ranking algorithms to calculate reputation and expertise along social graph composed of users’ links for information retrieval, to promote experts and demote spams. However, these existing works shed little light on personalized expertise ranking algorithm from view of the topic-relevance between users’ interests.In this study, we demonstrated the existence of homophily in users’ interests measured by social annotations using AskMeFi, a large scale community driven question and answering (CQA) system. Based on our investigations, we proposed Human-centric Topic Relevance based Personalized Expertise Ranking, a graph-based algorithm which takes the topic-relevance and authority among co-interest users and time traits into the computation of the expertise level of users. To demonstrate the feasibility and algorithmic performance of our proposed topic relevance based personalized expertise ranking algorithm, we implemented our experiments on the dataset from CQA website AskMeFi. The experimental results revealed that our proposed algorithm significantly outperforms other non-personalized expertise ranking algorithms.