Research On Answer Summarization Method In Question-answering Community By Integrating Answerer Ranking Score

Baidu Knowledge,Zhihu,Tianya Q&A,Quora,and other Q&A communities meet the needs of users to share knowledge,where users act as both question-askers and answerers.The answer summarization method integrates multiple answers under the question to get a concise and correct answer summarization,avoiding users spending a lot of time and energy to obtain useful information from multiple answers.The expert discovery method identifies potential high-quality respondents to new questions,which is conducive to ensuring that the questioners can receive timely responses.Using the expert discovery method to obtain the ranking of multiple respondents under the question and integrate it into the answer summarization process can help the answer summarization method find high-quality answers and improve the quality of the answer summarization.The main work of this paper is as follows:1.Expert discovery method of Q&A community based on interest and Expertise modelingExisting expert discovery methods in Q&A communities typically model user interests based on one-directional information learned from the sequence of questions that users have answered while ignoring the fluctuation of user interests and the impact of sparse behavioral data on the accuracy of user interest modeling.In addition,they do not consider the role of semantic correlation between historical answers and questions in evaluating user performance.So,a Q&A community expert discovery method based on interest and expertise modeling is proposed.First,BERT4 Rec is used to learn the two-way information of the user’s recently answered question sequence to obtain the recent dynamic interest representation;Secondly,build the user social network,use Deep Walk algorithm to learn network structure features,get the long-term interest of the user representation,and eliminate the impact of data sparsity on interest modeling;Then,a user expertise evaluation network is built to calculate weights for corresponding questions based on the semantic relevance between user answers and questions,as well as feedback information.The attention mechanism is introduced to focus on users’ performance on questions that are similar to the new question,resulting in a user expertise representation.Finally,the dynamic interest,long-term interest,and expertise representations are integrated and matched with the new question for scoring to identify users who are willing to accept invitations and can provide high-quality answers.The experiment shows that this method has achieved good performance.Compared with the baseline method,the MRR evaluation index of English,3dprinting,and Tianya Q&A data sets have increased by 5.2%,2.7%,and 16.1% respectively.2.Answer summarization method of Q&A community integrating respondents ranking scoresIn response to the insufficient modeling of sentences and the neglect of the role of answerer-related information in the answer summarization process,a new approach for summarizing answers in a question-answering community is proposed.Firstly,Ro BERTa-wwm is used with average pooling to encode the sentences and obtain their representations.Then,the relevance score between each answer sentence and the question,the novelty score of each answer sentence,and the ranking score of the answerer are calculated based on a previously trained expert discovery model.Finally,the sentence’s final score is determined by integrating its relevance,novelty,and ranking scores,and the MMR algorithm is used to iteratively select sentences to construct the summary.The experimental results show that using Ro BERTa-wwm with average pooling can better capture the deep semantic information of answer sentences.The integration of the three evaluation scores considers the interaction between answers and questions as well as the interaction between answers and answers,while also effectively incorporating information about the answerer,thus improving the quality of answer summarization.3.Q&A community answer summarization prototype system integrating respondents’ ranking scoresA prototype system for answer summarization was designed and implemented based on Python language and Flask framework.The system has designed three modules: data preprocessing,expert discovery,and answer summarization,which fully restore the research method of this article and visually display the answer summarization results.