Research On Key Technologies Of Event Graph Construction In Tourism Field

According to statistics,the number of domestic tourists reached 3.246 billion in 2021,and the total consumption reached 2.92 trillion yuan.Such large-scale tourism consumption demand is a huge challenge to the existing Internet platform.How to provide dynamic,personalized and intelligent services for tourists is a demand that cannot be ignored.Most of the traditional tourism recommendation systems only consider the static knowledge in the tourism field.However,the spatial transfer of tourists and the tourism events participated by tourists have not been considered.As an indispensable infrastructure of the new generation of cognitive artificial intelligence,event graph can better describe the essence of the world and provide a solution to improve the dynamic,personalized and intelligent online tourism services.According to the industrial needs of the tourism vertical field,this thesis studies the key technologies of the construction of the event graph,and puts forward an implementation scheme of the automatic construction of the event graph in the tourism vertical field.The system is composed of event extraction framework based on unstructured text data of domestic tourism websites,event temporal relationship extraction framework,knowledge representation and storage modules.The main contributions and innovations of this thesis are as follows:An event extraction algorithm based on machine reading comprehension is proposed.The event extraction algorithm models the event extraction task into the form of multiple rounds of QA.Compared with the traditional sequence labeling algorithm,the scheme can integrate the semantic information of the tag,so that the model has enough prior knowledge,so as to better identify and extract events.The experimental results of the event extraction algorithm on English public dataset ACE2005 and Chinese dataset DuEE far exceed the baseline model.This thesis puts forward the information extraction framework of tourism vertical field.The extraction framework first defines the event types of tourism vertical field,and then constructs the data sets of tourism vertical field event extraction and tourism event temporal relationship extraction based on these event types.Furthermore,an event detection scheme based on attention mechanism is designed for the dataset,which can automatically identify the event trigger words in the text.Finally,a temporal relationship extraction scheme of tourism events considering efficiency and performance is proposed.This scheme can still ensure the high recall and precision of relationship extraction in the text containing only a small number of events.This thesis constructs a travel vertical domain event graph for travel notes data named TravelEG.Exploring the integration of knowledge graph and event graph,considering dynamic information and static knowledge,can more comprehensively describe the journey of tourists.The graph contains about 2.9w+Event nodes and 8W+relationships,which can provide a professional knowledge engine for the needs of tourism intelligent recommendation in the industry.