Research On Social Question Answering Community User Answering Behavior Prediction Based On Recurrent Neural Networks

Web2.0 is the core concept of the Internet age.The progress of social science and technology enlarges people's knowledge needs.Users are no longer confined to the first generation of "keyword search" networks question-and-answer system in the process of seeking answers.The socialized question-and-answer community has become the preferred way for users to "ask questions and solve problems".Social Q&A community has the characteristics of sociality,interaction and sharing.It can produce a variety of high-quality answers to a question.Question users screen the answers again in the process of browsing,which also promotes the emergence of community opinion leaders.Opinion leaders,on the one hand,use the Q&A community platform to ask questions and acquire the knowledge needed by individuals;on the other hand,in order to achieve self-worth and balance the psychological needs of respected and needed,they will share more information to meet the knowledge needs of other users.Based on the links of social networks relationships,the satisfaction of personal knowledge,and the need to realize the self-worth of community opinion leaders,more users are attracted to form stable social networks relationships,resulting in information behaviors such as questions,answers and comments.Because the questions answered by users are often directly related to users' interests,that is to say,the generation of user's answering behavior is based on users' interests.Only when users are interested in a certain kind of questions,can they answer such questions.Based on the relationship between user's interest and the question answered,this paper starts with the question answered by the user,realizes the recognition of the question topic by constructing the question topic prediction model,predicts the probability of the question topic according to the model,and combines the probability of the user answering a topic,thus predicts the probability of the user answering a question.At the same time,there are question labels in all the questions answered by users,and the question labels can be used as the matching basis of question categories(topics).So we can predict the probability of a user answering a question from the perspective of the question he or she answers.The work of this paper mainly includes the following aspects.Designing and generating the data set needed to construct the recurrent neural network model: Combining the existing prediction method of microblog forwarding and the method of using neural network to predict the click-through rate of advertisements in many fields,by analyzing the users and the characteristics of social question answering community,using the crawler technology to collect the relevant points of social question answering community users and questions on the Internet.The collected data are organized into data that can be received and processed by the recurrent neural network through the eigenvector method.Designing and constructing the prediction model of social question-answering community user's answering behavior: To realize the prediction of social questionanswering community user's answering behavior,first of all,we need to establish a problem topic prediction model based on recurrent neural network(RNN)network architecture,according to the forward and backward propagation algorithm of RNN,configure the development environment and use the TensorFlow framework to configure the nerve.Network implementation.According to the collected topic and problem related data,the feature of question content and problem label are selected as input of the model,and the problem topic prediction model based on recurrent neural network is constructed.According to the output of the model,the topic and probability value of the question can be obtained,and the probability of the user answering a question can be predicted by combining the probability of the user answering a topic.Through experiments,the parameters of the model are optimized and the performance test is carried out.The model is used to learn the questions answered by the users in the social Q&A community.The parameters of the model are continuously optimized through experiments,and the optimal prediction model is obtained.Then the problem topic and its probability are predicted by the model.On this basis,three problems and prediction results are selected,and the effectiveness of the RNN prediction model is proved from three aspects: whether the prediction results of the recurrent neural network model conform to the characteristics of its output layer Softmax function,whether they conform to the retrieval results and the problem/topic mapping relationship.The model is evaluated by combining the learning rate and the defect rate index.In this paper,from the point of view of the questions answered by users,the recurrent neural network model is used to predict the behavior of answering questions,and the topic prediction model based on recurrent neural network is built.Combining with the topic answering probability of users,the question answering probability of users is predicted.This topic can not only fully tap the key characteristics of users and problems,but also provide a certain research basis for fine-grained prediction in other fields.It also has great research value for the recommendation of interest issues,the discovery of hot-selling commodities and the implementation of commercial marketing strategies.Although this paper predicts the answer probability of social QA community users by using machine learning algorithm to construct a topic prediction model,which has certain theoretical value and practical significance,some techniques of in-depth learning still need to be optimized and perfected.