Research On The Ranking Of Social Q&A Community Answers Based On Multidimensional Features

In the Web2.0 environment,the social Q&A community has become the largest knowledge platform for users to obtain information and interact with each other.While the Q&A community brings convenience to users,it inevitably leads to the problems of"information overflow" and "information overload".The existing social Q&A community provides the mechanism of answer filtering and answer ranking to provide users with high-quality answers,but there are some problems such as single answer ranking.Therefore,it has profound practical significance to construct a social Q&A community answer ranking model.The existing research mainly proposes a new ranking algorithm,ignoring the construction of the feature system of answer ranking,the distinction of feature importance and the evaluation of different types of ranking learning algorithms.Therefore,this paper constructs a social Q&A community answer ranking feature system from the multiple dimensions of answer,writer and voter features,compares the applicability of 11 ranking learning algorithms based on deep learning,tree,neural network,support vector machine and so on in Q&A community data set,and trains random forest classification algorithm to get the importance of each feature.To evaluate the ranking methods and features of social Q&A community answer ranking tasks.This paper mainly includes three aspects:Firstly,we construct the answer ranking feature system of the multidimensional features of the social Q&A community.The initial system is proposed from the dimensions of answer,writer and voter features.Python 3.6 and scrapy were used to build the crawler,we collect relevant data such as questions and answers,and use natural language processing technology to quantify the features.According to Pearson correlation coefficient matrix and the degree of contribution of high correlation characteristics to model fitting,the ranking feature system of social Q&A community is determined.Secondly,we compare the applicability of different types of ranking learning algorithms on social QA community data sets.In this paper,the open source ranking algorithm toolkits SVMRank,RankLib and TF-Ranking are used to test 11 ranking algorithms with 10 fold cross-validation,and the performance differences of ranking algorithms on evaluation indexes(NDCG@k,MRR)are compared.The experimental results show that the ranking algorithm based on deep learning performs best,followed by the random forest ranking algorithm.Third,we compare the importance of the ranking features of answers from different dimensions.This paper calculates the importance of different ranking features by using the random forest classification algorithm in the python machine learning algorithm library scikit-learn,and obtains the importance results of features according to the feature_importances_variable.The experimental results show that the features of voters' influence are very important,followed by the features of answer content,and finally the features of respondents'speciality.