The Design And Implementation Of Community Detection Algorithm For CQA Systems

Recently,Community Question Answering(CQA)has gained more and more scholars' attention.However,the majority of these research mainly focused on the question and answer(Q&A)content,there are still little research on the user network of CQA.In the domain of complex network,community detection is a hot research topic.There are three main reasons why community detection algorithm that works well in traditional social networks can not be migrated directly to CQA.Firstly,CQA system has a different organization structure compared with traditional social networks.The structure of CQA system is tree-shaped,while the structure of traditional social network is flat and non-hierarchical.Secondly,the ways users interacting with each other is different.There is no explicit connection between CQA users,whereas users in traditional social networks are directly connected.Furthermore,CQA has a different network from traditional social network.All entities and edges of CQA form a complex heterogeneous network,while traditional social networks are homogeneous networks.This paper is aiming at constructing proper networks according to the features of CQA systems and designing more applicable community detection algorithms.The main contributions of this paper are as follows:1.We propose a way to build user homogeneous network,and propose a hierarchical and overlapping community detection algorithm LPBHO.LPBHO puts forward a method to measure the node influence which take both the global information and local information into consideration.Based on the node influence,LPBHO puts forward influence propagation factor and integrates the above two points into node updating order and label selection strategy to complete overlapping community detection.After this,LPBHO completes hierarchical community detection through constructing overlay graph.2.We propose a way to build question-user bipartite network and propose an overlapping community detection algorithm BiHU.In oder to improve the poor performance of LPAb,this paper makes improvements from the following three aspects updating order,updating method and label selection strategy.The node importance is proposed as the updating order and the label list is set up to complete the hybrid updating.By adding the core degree to a community and edge density contribution of a node,BiHU algorithm is implemented.3.To solve the time-consuming problem of BiHU,we proposed parallel-BiHU based on Hadoop framework.Comparing with BiHU on experimental datasets,we verify the time performance of parallel-BiHU.