Research On The Method Of Discovering Experts In Q&A Communities Based On Topic Model

With the rapid development of the Internet,the Q&A community has become an emerging way for users to solve problems.However,in the Q&A community,there is a widespread phenomenon that users wait for the answer to be replied too long,and questioners cannot find domain experts in the community accurately and efficiently.How to recommend the domain experts of the Q&A community to the questioners who raised the questions has become a difficult point in the Q&A community algorithm.The mainstream methods for finding Q&A community experts mainly include webbased methods and topic model-based methods.The purpose is to model the user's professional field and professional level based on the text generated by the user in the community or according to the relationship between the users,and then accurately recommend the questions raised by the user to the domain experts of the Q&A community according to the calculated model.However,for most Q&A communities,the scale is relatively small,the text generated by the user is short,and the features are sparse.The traditional topic-based expert discovery method has limitations in mining short texts generated by users.Therefore,this paper analyzes the difficulties of today's Q&A community expert finding methods,and integrates many subject areas such as natural language processing,machine learning,and decision theory,and conducts research on the method of finding community experts in Q&A community.The main contents of this paper are as follows:(1)Improved algorithm based on Meta LDA(Meta LDA Topic Model).This paper aims at the existing research methods based on the topic model.When discovering the professional fields of experts in the question-and-answer community,due to the sparse text features generated by users,the problem of the method is not effective in finding the professional fields of experts.Algorithms are found in the professional field of Q&A community experts.First,extract the question-answer pair text and the corresponding metadata from the question-and-answer data of the community,and then use the Meta LDA algorithm to perform text clustering on the text of each question-answer pair to obtain the topic distribution matrix of the question-and-answer pair text.Then connect the topic distribution matrix of the question and answer text to the sentence vector of the question and answer text obtained by training using BERT(Bidirectional Encoder Representation from Transformers).Using the automatic encoder for feature extraction,cluster with Kmeans to get the clustering tags adding semantic grammar and topic distribution features.Use the calculated cluster tags,the pre-processed question and answer pair text,and the answerer,timestamp,the topic tags of question and answer pair text,question and answer pair ID as document-level metadata and word-level metadata processed by Glo Ve,the word embedding tool.The user's professional field is modeled with the improved Meta LDA algorithm.Using Gibbs sampling method to infer the topic parameters of the model,so as to count the user's professional field.Finally,combined with the praise information of the answer,the user's professional level distribution is calculated.Moreover,the method based on the topic model utilizes the co-occurrence feature of the lexical items in the text when clustering,without taking into account the semantic grammar and topic distribution feature added with context information.Finally,this article uses the question and answer data of the small elephant question and answer community as a data set to verify the effectiveness of the algorithm.(2)Combined with the user's multi-dimensional data,a multi-attribute decisionmaking method is used to measure the professional level of experts in the question and answer community.Aiming at the limitation of the existing network-based expert level measurement method: only a certain relationship between users can be modeled,and the multi-dimensional data of question and answer community users cannot be well used to measure the user expert level.A professional level measurement method based on multiattribute decision-making is proposed.This article uses multi-dimensional data from the Q&A community experts to conduct experiments.The data includes the user's social attributes and other behavior data related to the Q&A behavior generated by the user in the Q&A community.The algorithm first uses the maximum deviation method in multiattribute decision-making to determine the weight of the attribute,and then uses the multiattribute decision-making method to rank the influence of experts.Finally,use the question and answer data of the small elephant question and answer community as a data set for comparative experiments.Compared with the experimental results of traditional expert discovery methods,the experiments verify the effectiveness of the algorithm proposed in this paper in evaluating the professional level of experts in the question-andanswer community.(3)It combines the improved algorithm based on Meta LDA and the multi-attribute decision-making method.Use Meta LDA's improved algorithm to calculate the user's professional level distribution in various fields,and use the multi-attribute decision method to calculate the user's overall professional level as a weight,measure the professional level of the question and answer community experts in each field,and thus discover each Expert users in the field.