Research On The Quality Evaluation Of Answers In Social Q&A Websites

In the context of the Web2.0,with the rapid development of social networking sites and self-media,the social question-and-answer website represented by Zhihu and Quora gradually replaced the traditional search-based Q&A website.Compared with traditional search-based question-and-answer platforms like Baidu Zhidao and Sina Ask,social Q&A websites have the basic characteristics of “social relationship” and “question-answer mechanism”.They usually do not rely on an authority or expert to provide information,but use UGC(User Generated Content),each user has the triple role of information receiver,information provider and communicator.Because the social question and answer website encourages everyone to participate in the content creation and sharing characteristics,the quality of the website content will inevitably be affected.In order to help the website to select high-quality content for push,to ensure that the high-quality content of the website is properly disseminated,and to limit and improve the low-quality content,the evaluation of the quality of the answer is essential.This thesis mainly studies the quality evaluation method of answers in the social question-answer platform.Based on information adoption model,this thesis combs and summarizes the research status at home and abroad,and combines with the analysis of the data of Zhihu,chooses the surface attributes of the answers,the text attributes of the answers,the influence attributes of the respondents and the professional attributes of the respondents.as the initial evaluation features.Using web crawler to capture 9484 "question-answer-respondent" information that is representative of the online representative,through the mathematical statistics,machine learning,emotional lexicon-based text sentiment value calculation and other methods to extract the relevant characteristics of the answer quality evaluation,Use feature engineering related methods to filter features,and obtain a Index system including 13 index such as Text Length,Average Sentence Length etc.By comparing the experimental results of several classification algorithms,the best random forest algorithm was selected to construct the model.The experimental results of the final model on the test set showed that the classification accuracy of the model reached 79.6%.The effect of the feature of the respondent on the quality of the answer is verified by substituting the feature set into the model and analyzing the importance weights of each feature in the model.The results of this thesis can provide an idea for the social question and answer website to optimize the answer quality assessment method,and help the social question and answer website to improve the user experience.