| With the rapid development of the Internet,tourism has ushered in new changes in smart tourism and digital tourism.The method of tourists obtaining scenic spot information through traditional search engines is not direct and efficient,and the different contents of different websites and false advertisements are increased in disguise,which further aggravates the travel threshold burden of users.Therefore,providing users with an accurate,convenient,real and effective tourism information retrieval service can reduce the time cost for tourists to obtain information.The emergence of knowledge graph provides an effective solution for standardizing massive data.Q&A model based on knowledge graph has outstanding advantages in question sentence understanding and answer display.Taking Shandong Province as an example,this thesis constructs the knowledge graph of tourism field by using the powerful semantic pre-training model ERNIE,and uses it as the knowledge support to design and implement the tourism intelligent question answering system.The main work of this thesis is as follows:(1)Propose and achieve a set of knowledge graph construction process in the field of tourism.This thesis collects semi-structured data and structured data with strong objectivity from major tourism websites;Improve the ontology construction method,based on the seven step method,combined with the data characteristics and graph application requirements in the tourism field,and build a highly inclusive tourism ontology;In order to solve the problem of heterogeneous data fusion,this thesis proposes an entity alignment model POI-ERNIE-SN,which deeply combines the geographical location characteristics of entities.Multiple groups of control experiments are designed to prove the effectiveness of the model;In this thesis,Jena and Fuseki are used for knowledge reasoning and Baidu map API is used for knowledge expansion to solve the problem of some isolated nodes;Finally,the graph database Neo4 j is used to store knowledge and construct the tourism knowledge graph of Shandong Province.(2)The mainstream rule-based Q&A has the problems of high labor cost and poor mobility.This thesis takes the QA model based on deep learning as the main body,supplemented by a small number of template rules.The overall question answering model mainly includes four tasks: question entity recognition,entity chain reference,attribute classification and attribute fine sorting.For the recognition of core entities in question sentences,a named entity recognition model ERNIE-CRF is designed;An entity chain index algorithm based on fuzzy query is proposed;Design and improve the attribute classification model ERNIE-CNN,which integrates answer information.The coding layer of ERNIE word vector is followed by a layer of CNN to further mine the deep text features,and improve the model input based on the idea of hop by hop,so that it can deal with complex two hop questions;An attribute fine sorting algorithm based on combinatorial similarity is proposed.Finally,the feasibility and superiority of the overall QA model are proved by experiments.(3)Based on the knowledge graph and Q&A model,this thesis designs and implements an intelligent Q&A system in the tourism field of Shandong Province,which can deal with the problems in the graph,weather Q&A and chat dialogue.Due to the serious impact of COVID-19 on local tourism,the system adds the epidemic prevention and control query function on the basis of the existing QA system,which is more suitable for tourists’ travel needs.In summary,this thesis applies the semantic pre-training model ERNIE to construct the knowledge graph of tourism field,designs and improves the Q&A model supplemented by template rules and based on in-depth learning,and designs and implements the tourism question and answer prototype system based on knowledge graph.The results of this work will reduce the time cost for users to obtain tourism information and provide support for promoting smart tourism and digital tourism. |
PDF Full Text Request