Research On Interactive Question And Answer Technology Based On Deep Learning

With the development of society and the arrival of the Internet era,intelligent interactive system has been widely used in daily life and work,such as intelligent customer service and intelligent speaker.Interactive q?A refers to the feedback of user demand information in the context of human-computer interaction,in which the key problem is how to learn and represent the above information to make up for the semantic deficiency of the current discourse.In recent years,deep learning has proven its powerful feature extraction and representation capabilities in various NLP tasks.In view of this,this paper adopts deep learning to model the context information in interactive scenes,and the research work is summarized as follows:Existing intention recognition model is generally based on context-free assumptions,without considering the context dependencies between,and the existing memory network can't realize the above access to key information,in order to solve this problem,put forward the intention recognition method based on attention memory network,the experimental results show that this model can effectively capture the above key information,accuracy was 1.3%higher than the baseline model accuracy.Discourse rewriting in interactive scene can solve the phenomenon of reference and omission to a certain extent,but the existing context rewriting network model has the problem of decoding end repetition.To solve this problem,the Seq2Seq+Patt model was used in this paper to solve the repetitive decoding problem.Bleu improved 0.5%compared with the baseline model.Answer matching is one of the key tasks in the interactive Q?A system.Due to the absence of referential omission in human-machine interactive dialogues,it is more difficult to select answers.In view of this,this paper proposes an interactive scenario-oriented answer matching method,introduces the above information through "gating mechanism" to make up for the semantic information missing in the interactive scenario questions,and constructs the feature representation of the answer according to the questions through attention mechanism to learn the semantic matching relation of text fine granularity.First,collect the goods from the microblogging site review corpus,and on this basis to establish an interactive q?a corpus,and puts forward two kinds of gating and attention mechanism based neural network the answer match method,the experimental results show that the proposed answers matching method can effectively capture the information above,2.8%higher than the baseline model accuracy.In the end,we design and implement an interactive question and answer recognition and answer matching system,which can identify the input above information and the current user's utterance,and match the current questions and candidate answers.