Fusion Method Explicit And Implicit Hierarchy Relationship Expert Network Construction

Currently, expert search is the most popular field of vertical information retrieval domain. It is currently the most effective means to get expert information. While there are complicated relationships between experts and experts, this relationship network is a very important resource to support expert search. The reliability of the expert networks directly affects the accuracy of the expert search results. Expert relations include both explicit relations and implicit relations. The explicit relations are the relation between basic information of expert homepages. The implicit relations are the implicit semantic relations. Therefore, this thesis focuses on extracting expert explicit relations, extracting implicit subject relations and researching on how to integrate the explicit and implicit relation to construct expert relationships network. The emphasis achievements are as follows:(1)Extract two types of explicit relations according to the co-occurrence of organizations. Firstly, aiming at the problem that requiring large amounts of labeled training data while using supervised learning to extract the expert metadata, we propose a semi-supervised learning algorithm based on co-training style, identify all five types of metadata items from expert home page and decrease the manual markers. Secondly, we define expert-organization relationship rules, and extract the relations between expert and organization by rules matching. Finally, two types of explicit relations are extracted according to the relations of organizations, and expert explicit relation strength is computed by the percentage of co-occurrence metadata.(2)Extract implicit topic relations based on the ES-LDA model. First of all, according to the characteristics of the expert home page, there are four features, including expert metadata, metadata relations, home page links, as well as metadata co-occurrence. Secondly, we use the above four features guide traditional LDA. Then a supervised topic model ES-LDA is proposed to analyze the expert home-page themes. Finally, implicit topic relations are extracted based on the same topic of different experts, and expert implicit relation strength is computed by the percentage of the same number in high probability and the difference of document-topic distribution.(3)Integrate explicit and implicit relations to build experts relationship networks. Firstly, the total relationship forms are the sum of explicit and implicit relationship forms. Secondly, the total relationship strength is the weighted average of the explicit relationship strength and implicit relations strength. Finally, we respectively build expert relations network with experts as node, expert relations form as edge, expert relations strength as weight.(4) Build the prototype system of expert level relationship network with integrating explicit and implicit relations. The network provides important support for expert search.