Expert Discovery Method For Question Answering Community Based On LSTM Model

With the continuous development of the web2.0,More and more CQA(Community Question Answering)platforms are coming into people's lives,and more and more people tend to share their questions and answers through the CQA platform due to it's openness and sharing.Today,the CQA platform has become an important medium for users to access information and share knowledge.The CQA platform,such as Baidu know,Zhihu and Yahoo! Answers will publish a large number of questions every day which belong different topics,therefore,we should require the CQA system to have sufficient data resources and powerful problem-solving mechanisms to help users get highquality answers for their questions quickly and efficiently.However,with the rapid development of the CQA platform,the amount of platform data is expanding rapidly,and the massive platform information will quickly cover the questions of the questioner,as a result,the questioner has to wait for the answers of other users for a long time.At the same time,coupled with the interference of a large amount of spam and low-quality answer information,it is difficult for users to obtain high-quality answers quickly and accurately for their questions,which increases the great pressure for users to obtain answers.In view of the above problems,this paper research the expert discovery method in the CQA platform systematically,which is aimed at discovering users with relevant expertise or experience for a given question.The main work include the following aspects:(1)Based on the user history question and answer information,this paper proposes the topic professional level model TPLM,which uses the tag information,voting information and time information of the post to model the user.Firstly,the tag information is added to the user document,and the LDA model is used to model the user document.Then,based on the topic probability distribution,the voting information of the post is used to model the professional level,and the user's professional ability under each topic is evaluated.So as to better mine expert users with strong professional skills under relevant topics.(2)Based on the calculation results of the topic professional level model(TPML),this paper uses the TPLMRank method(the fusion method of TPML and PageRank)to evaluate the user authority.Based on the TPLM model,the method integrates the link structure information of the user QA(question-and-answer)relationship network,and integrates the TPLM model with the PageRank algorithm.First,the user history QA relationship is used to form a QA relationship network between users,and then the calculation result of TPLM is used as the dependency factor of the jump between user nodes in the PageRank algorithm.Finally,the user is comprehensively scored based on the TPLMRank method.(3)In view of the problems in the platform,when calculating the relevance of users and problems,this paper improves the similarity calculation method between user and problem by takes into account the topic information and deep semantic information of the text,and uses the SLALDA model(the fusion model of LDA model and Siamese LSTM model based on Attention mechanism)to replace the original LDA model.It first uses the LDA model to calculate the similarity between the user and the problem,and then uses the Siamese LSTM model based on the Attention mechanism to calculate the deep semantic similarity between the user and the problem.On this basis,the two are merged to obtain the final semantic similarity between the user and the newly asked question.Finally,based on the user authority obtained by the TPLMRank method and the similarity between the user and question obtained by SLA-LDA,the SL-TPLMRank method is used to comprehensively score and sort for users,and recommend the top users to the questioners.The experiments in this paper were carried out on the real corpus of the Zhihu QA platform,and the experimental results were evaluated using NDCG and MRR.The evaluation results show that the semantic analysis technology based on TPLM model can effectively mine the interest distribution and professional competence of experts.The TPLMRank algorithm based on TPLM model can more accurately measure the authority of users.The model who integrates LDA and Siamese LSTM based on Attention mechanism can also better mine the semantic feature information of text.In general,the relevant models and methods proposed in this paper can improve the quality of expert discovery to a certain extent.