Answer Generation Assisted By Review Data In E-Commerce Question-Answering

With the development of the Internet 5G technology and the increasing demand for online shopping,in order to enhance the user's shopping experience and promote the growth of the platform's economic benefits,research on intelligent question and answering of products in ecommerce scenarios has begun to attract attention.On the one hand,it is still a challenging task to design an intelligent question-answering system for products in e-commerce platforms due to the limitations of technology,the amount of data,and high noise in the raw data.On the other hand,graph-structured networks have a natural interpretability advantage in representing unstructured text information.Graph-based neural network methods have begun to attract the attention of many scholars.In addition,the development of sequence-to-sequence encoding and decoding technology has established a research foundation for the generative intelligent question and answering.Due to the requirements of practical application and the technical feasibility,we proposed an intelligent question and answering model RDQA for e-commerce filed based on product review data.Overall,the RDQA model we proposed in this paper is based on the sequence-to-sequence architecture,taking the questions raised by the user in the e-commerce platform as input,the answer to this question as the output of the model,and the review data of this product used as the external knowledge to supply the answer generation.Specifically,RDQA extract valuable knowledge information to form an unstructured graph network from product reviews based on the semantic similarity between question and reviews.And more,the RDQA model uses the Transformer network as the basic unit of sequence-to-sequence architecture,and uses the semantic encoding of the input sequence to update the representation of the graph node through the attention mechanism.In process of the answer generation,the prediction of the next word can be accomplished by dynamically updating of the hidden state in the generation process based on the semantic encoding of the input question and the representation of related graph node.Because the user-generated content is a little colloquial in e-commerce platform,the RDQA model introduces a pointer generator network mechanism in the prediction process in order to avoid too many repetitive words in the generated answers,giving the model the ability to select words from the input question as output words.We compare the RDQA method with some state of the art neural answer generation model in this paper.The experimental results on two public real e-commerce platform datasets both show that the RDQA model is superior to existing neural answer generation methods in generating product-aware answer in e-commerce field.For further research,we will continue to optimize the knowledge extraction of product reviews according to the user intent,and try to use the social information such as the similarity of shopping habits and preference between different users to drive the further development of product-related intelligent question answering system in e-commerce.