Multi-Source Event-Centric Knowledge Graph Fusion In Tourism

Spatial and temporal information plays an important role in urban computing,tourism planning and other applications.Therefore,in the field of smart tourism,we hope to mine the spatio-temporal trajectory data of tourists for better analyzation and application.At present,there are many sources of tourist spatio-temporal trajectory data,such as location information recorded by mobile devices and user-generated content data,which have their own advantages and disadvantages in location and semantic accuracy.Therefore,we hope to make up for the incompleteness of a single source by fusing spatio-temporal trajectory information from multiple sources and better serve various smart tourism applications.However,it is difficult to use the traditional data fusion method because of the large gap in the structure and representation of spatio-temporal data from different sources.Therefore,we regard the behavior of tourists as the sequential transfer of tourism events in time and space,with the help of event-centric knowledge graph,an extension of the knowledge graph describing events and events transfer,to describe the trajectory of time and space behavior of tourists.So we convert the spatio-temporal data from different sources into the spatio-temporal event-centric knowledge graph,so as to realize the fusion of multi-source knowledge through the way of the knowledge graph fusion.Compared with the traditional knowledge graph,the integrated spatiotemporal event-centric knowledge graph contains more precise semantic and temporal information,which is of great significance to various applications in tourism scenarios.Based on the spatio-temporal eventcentric knowledge graph after fusion,this paper studies the recommendation algorithm of the next Point of Interest(POI)with spatiotemporal event-centric knowledge graph to solve important recommendation problems in travel scenarios.The main work and innovations of this paper are as follows:(1)We propose a framework for fusing multi-source spatio-temporal data by means of spatio-temporal event-centric knowledge graph.That is,the spatio-temporal event-centric knowledge graph is constructed from spatio-temporal data from different sources,and the purpose of knowledge graph fusion is achieved by means of graph embedding and entity alignment.At the same time,a long-path-based knowledge graph representation learning algorithm LPTransE was proposed to describe the spatio-temporal specificity of the constructed spatio-temporal eventcentric knowledge graph in the embedding part,that is,by means of attenuation factor and iterative learning path representation,the lowdimensional vector representation of entities and relations was learned while capturing the long relational path in the event-centric knowledge graph.(2)Based on multi-source spatio-temporal event-centric knowledge graph after fusion,we propose a personalized recommendation algorithm.On the basis of graph attention network,the specificity of relation path is considered to improve the calculation of attention score,and finally the multi-hop neighbor information in the graph is aggregated to enrich the item representation.At the same time,long and short term memory network is applied to capture the dynamic personality preferences of users by inputting user’s history records.Finally,the design and implementation of the recommendation system are completed in the real tourism scene.