Study Of Question Recommendation In Community Question Answering

With the rapid development of Web2.0,Personalization and communalization become the trend.Community Question Answering(CQA)site has become more and more popular for its flexible user interaction characteristics.At present,a lot of CQA websites have appeared on the Internet.These communities have accumulated a large number of users and content resources,and show a trend of continuous growth.The community is flooded with new questions every day,waiting to be answered by other users.A set of possible answerers should be recommended to shorten the waiting time of askers and help answerers find suitable questions to answer,thus improving user stickiness.The main content of this paper is to explore the method of mining the user's interest according to the user's answer records and then recommend possible answerers for new questions according to the user's interest.First of all,the tags of the question in the community contains a lot of valuable information.For each question,the label provided by the questioner can be regarded as the most representative word of the questioner's focus and intention.In this paper,we propose a question recommendation method based on time-sensitive tags,which calculates the user's score under each tag according to the number of tags in the user's history to answer questions,and then generates a user-tag matrix,and finds the respondents interested in the new questions according to the matrix.At the same time,the importance of different tags and the change of users' interest over time were considered.The experimental results show that this method is effective,and the recommendation effect can be further improved after taking into account such factors as label weight and user interest change.Secondly,there are semantic relations between multiple tags of the question.Convolutional neural network can effectively mine semantic information between text words,which has a significant effect in text classification.In this paper,the convolutional neural network is used to predict the answerers of the question by text classification,input the text and tags of the question,and output the probability of each candidate user to answer the question.In consideration of using both the question tag and the question text,this paper proposes to use the double convolution channel to isolate the tag from the text,so as to avoid the mutual interference between the text and the tag.Further,considering the gap between the problem tag and the positive word space,this paper proposes to adopt a two-layer convolutional neural network,first mapping the text word into the tag space,and then mining the user's interest in the tag space for problem recommendation.Experimental results show the effectiveness of the proposed method.Finally,this paper uses a learning to rank method to train a mixed model that integrates the results of multiple recommendation models by taking the prediction results of multiple models as the eigenvalues,so as to realize the complementary advantages of multiple models.Experimental results show that this method can further improve the recommendation effect.