Research And Application Of Key Technologies Of Community Question Answering

As a new way of Q & A and knowledge sharing mode,Community Question Answering has been widely used in various fields in recent years.Automated Community Question Answering can effectively improve the efficiency of Q & A and enhance the user's sense of experience.At present,there are many researches on question answering classification,answer quality evaluation and so on,but there are few researches on answer selection and similar question retrieval task in Community Question Answering.Therefore,based on the Community Q & A data set released by Sem Eval,this paper uses machine learning methods to carry out innovative research on these two problems,and puts forward relevant technical solutions.Firstly,A Context-Relevance Answer Selection Model CRASM based on recurrent neural network is proposed for answer selection of Community Question Answering.The model uses multi-dimensional word vector fusion representation layer,attention layer and non-linear full connection layer to complete the semantic matching modeling of question and answer pairs,and uses the recurrent neural network Bi LSTM to complete the semantic association modeling between answers.The rationality of the model structure was verified by ablation experiments.Compared with other models,Macro-F1 of this model is 58.33%,and performance of this model is better.Secondly,this paper proposes a similar question retrieval model FFIRM based on multi feature fusion for the similar question retrieval task of Community Question Answering.The model first calculates the relationship between the lexical features and semantic features of the query and candidate question sentences,and then inputs these relationship features into ranking SVM to comprehensively judge the similarity between the two sentences.The ablation experiment and contrast experiment show that the structure of the model is reasonable,and the retrieval effect of similar questions is better.On this basis,the comprehensive performance test of information retrieval algorithm and the model that performs better in sentence matching task is carried out.By balancing the calculation cost and model performance,a two-stage question retrieval algorithm combining TF-IDF model and FFIRM model is proposed.Finally,using the above answer selection algorithm and two-stage question retrieval algorithm,an intelligent question answering system is constructed,and the overall architecture of the system is designed and the specific function modules are realized.Through the functional test and performance test of the system,the effectiveness of the algorithm and the usability of the system are proved.