Research Of Answer Ranking Method Based On Weighted Keywords

VWith the rise of Internet Question-Answering communities,answer ranking and content filtering in Q&A forums get more and more attention and research.Compared with some widely used content filtering methods such as vote mechanism,answer ranking methods based on machine learning are more generic and opportune,and the ranking results based on these methods are more reliable.But existing answer ranking methods based on machine learning either extract features from answers and related questions,users and comments,then train ranking model with Learning to Rank methods to rank answers,or calculate the expertise value of answerer under a specific domain with link analysis methods to rank answers.All these methods consider each question as a separate entity when ranking their answers.But questions in Q&A forums are related with each other,similar questions may have been post before and solved.These solved questions could be used to rank answers of new questions.Based on the above facts,in this paper,we proposed an answer ranking method which uses solved questions in its ranking process.The main research contents are as follows:1)The method of measuring question similarity.In this paper,we proposed a new text representation with Word2vec,then use it to calculate the similarity of question texts.Experiments with existing text similarity measurement methods proved that our method achieved better results.2)The method of extracting keywords.In this paper,we improved the weight initialization and weight distribution of classic TextRank method with Word2vec.and proposed a new keywords extraction method,Experiment results show that our method is better than other existing keywords extraction methods.3)The answer ranking method based on weighted keywords.This method acquires similar questions with question similarity calculation method,then extracts keywords from answers of similar questions,and evaluate the importance of answering the question of these keywords.At last,we evaluate the quality of answers to be ranked with the extracted keywords and get the final ranking results.To further get better answer ranking results,we combined the above ranking method with the method based on feature extraction and the method based on link analysis to put forword a new answer ranking method.Experiments on datasets extracted from Stack Overflow and Yahoo! Answers proved the effectiveness of the new proposed answer ranking methods in this paper.In summary,we proposed a new question similarity measurement method and a new keywords extraction method in this paper,then proposed a new answer ranking method based on weighted keywords with the above two methods.Experiment results with existing answer ranking methods demonstrate the effectiveness of our method and it can improve the quality of contents and the efficiency of user's access to these contents in Q&A forums.