Entity Enhanced Automatic Answer Text Generation

In recent years,automatic question answering system is widely used in e-commerce and other fields.In the field of e-commerce,because goods cannot be directly observed and experienced as offline,users have more or less questions about goods,and intelligent automatic question answering system can greatly reduce the cost of browsing and filtering information for users,and provide users with real and effective answers.With the development of natural language processing technology,the answer text generation method of automatic question answering system has gradually developed from the answer text generation based on fixed template to the answer text generation based on natural language.Natural language answer text generation,through the model to understand the problem,and then simulate the human response to generate natural and smooth answers,involving natural language understanding and natural language generation two key technical points.Although natural language based answer text generation has been able to generate smooth sentences,there are also problems such as lack of information content and inaccurate answers.In order to solve the problem of insufficient text information in the question and answer community in the field of e-commerce,this paper uses entity data sets and commodity related information to form a graph network,analyzes and uses commodity related information from the dimension of the graph,so as to improve the quality of the generated text.This paper proposes an entity enhanced graph convolution network answer generation(EGAG)model based on entity enhanced graph convolution network.On the one hand,effective information in entity graph network is mined through graph convolution network module,and on the other hand,effective information in commodity questions,comments and description texts is captured through self attention module,interactive attention module and segmented attention module,Then the answer is generated from different perspectives through the multi perspective fusion module combined with the question.Compared with the latest model on Amazon data set,the graph convolution network answer text generation model based on entity enhancement achieves higher results on the evaluation indicators of text generation based on word overlap and word vector,and the generated answer is also more meaningful.Because the pre-training language model can achieve significant improvement in most natural language processing tasks,in order to make better use of entity information on the basis of the pre-training language model,this paper proposes an entity enhanced prompt learning answer generation(EPAG)model based on entity enhanced prompt learning.On the basis of T5 pre-training language model,by marking entities,And the learnable identifier is designed as the template of prompt word prompt question statement,comment statement,answer statement and entity.EPAG model can make full use of the common sense and reasoning ability of pre-training language model under large-scale data training,and fine tune the generated text through the prompt learning of entity information.Through comparative experiments and template experiments,this paper verifies that EPAG model can generate higher quality answers and the rationality of template design.