Deep Learning-based Book Guide Answering Robot Research

The advent of the Internet era has made online shopping a more efficient and convenient way to consume.With the huge amount of product information,how to quickly locate and select the right products for users and improve their consumption experience and consumption efficiency is a key technical point of concern for the industry.For most online merchandising fields,the huge amount and complex structure of product data makes it impossible for traditional search engines or customer service bot systems to accurately understand users' requirements and give them the answers they want in a concise and precise manner.In most P2 P platform shopping guide scenarios,users must also perform multiple iterations of demand description and classification positioning,or directly contact human customer service to get information about their ideal products.The paper proposes a deep learning-based book guide answering robot system by introducing a Knowledge Base Question Answering(KBQA)technology with book commodities as the research area and its key technology of intelligent shopping guide as the research objective.By inputting book-related questions to get relevant content responses,we aim to analyze the semantics of the questions from users' needs and clarify their intentions,and then quickly match the answers through the knowledge graph of book information.The main work of this paper is as follows.(1)Construction of book information knowledge graph.Book information crawling was carried out on authoritative data sites by building a crawler program.Asynchronous crawling technology are adopted to improve the crawler operation efficiency,and the data is cleaned and integrated after being obtained.The Schema layer of the book information knowledge graph is constructed by analyzing the data,the Neo4 j graph database is selected as the implementation platform of the book information knowledge graph,and the top-down automated fast construction and visualization of the knowledge graph is carried out by relying on py2 neo.(2)A named entity recognition model(RSBC)with fused features is proposed.A book domain dictionary was constructed,and lexical boundary features were extracted using the Soft Word approach.The pre-trained model Ro BERTa character-level embedding is combined with it.And the Bi LSTM model is used to incorporate directional information and perform annotation error correction by CRF layer to improve the accuracy of named entity recognition.A pretraining based convolutional intention recognition model is proposed to combine the traditional Text-CNN with the pre-training model,which makes use of BERT's good at extracting text features and compensates for Text-CNN's shortcoming of not being good at extracting long distance text features,and achieves some effect improvement.(3)A deep learning-based book guide answering robot system is designed and implemented.Based on the constructed book knowledge graph and the designed intelligent Q?A algorithm,the book guide answering bot system is implemented.The front-end and back-end development of the system is realized by using the framework of Flask,etc.to complete the book shopping guide business and book-related knowledge Q?A.