Research On Question Response Time Prediction Of Community Question Answering

Community Question and Answering(CQA)has become an important way for Internet users to acquire and publish knowledge.Long response time is a common problem in the existing CQA,which can reduce the user experience.Therefore,the prediction of question response time began to attract the attention of academia and industry.However,there are some problems with the existing response time prediction methods: Less consideration is given to the impact of expert users on response time prediction,the impact of unresolved problems is ignored on response time prediction and the lack of extracted characteristics in response time prediction.These problems seriously decrease the accuracy of existing response time prediction methods.In order to solve these problems,this paper proposes a question response time prediction method based on question and answer model.For the newly asked questions,firstly,finding the specific experts to the question by the expert finding method based on label cluster topic and weighted PageRank.Secondly,question model and the answerer model are constructed based on multi-factor,then using the multi-factor model matching based question recommendation method to find the candidates.Finally,the response time of question is calculated by question and answerer model matching based answerer's response time prediction method.Main research basics and contributions of this paper are as follows:1.This paper proposes an expert finding method based on label cluster topic and weighted PageRank.This method includes two parts.First,Label Cluster Latent Dirichlet Allocation(LCLDA)method is proposed to solve the overfitting problem of traditional topic model.Second,an expert ranking method based on Topic Sensitive Weighted PageRank(TSWPR)is proposed to solve the traditional experts sorting method of identifying the Abuse users as experts.Firstly,the label clusters are clustered by clustering the label data in the CQA,and the LC-LDA model is constructed by improved the traditional LDA model with label clusters.Then the LC-LDA model is used to classify questions and experts in the community.This method reduces the over-fitting phenomenon of short text classification of traditional topic models.Secondly,we use voting data and askers' satisfaction to measure the quality of the answer,and combine the result of LC-LDA to improve the traditional PageRank algorithm,then the topic sensitive weighted PageRank algorithm is obtained.Using TSWPR algorithm to calculate the professional level of askers in different areas,answerers are ranked according to the professional level and expert users are acquired in different areas.Finally,the LC-LDA model is used to perform expert classification experiments on the Stack Overflow data set,and the classification results are better than the traditional LDA model.2.In order to solve the problem of insufficient features in constructing the model for the existing recommendation method,this paper proposes a Multi-Factor Model Matching based Question Recommendation method(MFMMQR).Firstly,the Multi-factor based Question Answerer Model(MQAM)is constructed which includes the answerer model and the question model.The answerer model describes the respondents through the characteristics of answerer interest,answerer professional level,answerer activity and so on.The question model describes the question through various characteristics such as question category,question difficulty,question time and so on.Secondly,we use the model matching strategy based on Model Similarity and the model matching strategy based on Factorization Machine to match the question and answer model,calculate the matching degree of each answerer to the question as the willingness to answer the question,and recommend the question to the answerer with higher willingness.Finally,the results of question recommendation experiment on the Stack Overflow data set shows that the MFMMQR method proposed in this paper has higher recommendation accuracy than the methods of extracting only a few Q & A behavior characteristics.3.In order to solve the problem of insufficient features in constructing the model for the existing question response time prediction method,this paper proposed a Question and Answerer Model Matching based Answerer's Response Time Prediction method(QAMM-ARTP).Firstly,this paper takes into account the answerer interest,answerer professional level,answerer activity and other respondent associated characteristics,and the question category,question difficulty,question time and other problem associated features of Q&A process,then the QAMM model is used to describe the answerer model and question model.Secondly,the response time of each user is calculated by matching the Q&A model with the model matching strategy based on SoftMax classifier.Finally,it is proved that the QAMM-ARTP method proposed in this paper is more accurate than the traditional response time prediction method based on the question-related feature by experimenting with the question response time prediction on the Stack Overflow data set.