Research And Application On Automated Construction Of Knowledge Graph For Smart Trip

As socialism with Chinese characteristic has crossed the threshold into a new era,the principal consumption demand of people has evolved.What people mainly consume has gradually shifted from basic material,which satisfies material needs of the people,to tourism consumption,which satisfies the people's ever-going needs for a better life.Even under the shadow of COVID-19,the number of domestic tourists in 2020 reached 2.87 billion.Such a large amount of tourism consumption demand poses a great challenge for Online Travel Agent,such as Ctrip,to provide more personalized and intelligent online tourism services.Knowledge Graph,as one of the most important infrastructure for the next generation of cognitive artificial intelligence,has provided a solution to enhance intelligent ability of online tourism services.Based on industrial demand and research value of tourism domain,the thesis implement a smart trip knowledge graph system,SmarttripKG,which automatically construct a tourism domain knowledge graph from scratch.The SmarttripKG consists of Ctrip semi-structured knowledge extraction framework,distantly supervised relation extraction model DSK-DISTRE,tourism knowledge representation layer and other functional modules.The main contributions and innovations of this thesis are listed as follows:(1)To obtain billions of high-quality tourism knowledge triples and restrict system complexity,this thesis proposes a semi-structured knowledge extraction framework solely based on Ctrip,which extends DBpedia extraction framework and migrates it into tourism vertical domain.This framework extracts all kind of tourism knowledge,including restaurant,hotel,sight,shopping,place,from several semi-structured data source of Ctrip travel page,such as infobox,detailbox and HotelAPI.As a result,this proposed framework has become the main knowledge source of SmarttripKG.(2)To make use of unstructured travel notes,and extract triples from them,this thesis proposes a new distantly supervised relation extraction algorithm DSK-DISTRE,which achieves both high accuracy and high recall.Our model is experimented on the public dataset Riedel NYT10,and result outperforms baseline models,achieving AUC 0.448.(3)Based on the needs of tourism industry,this thesis proposes and open-sources a smart trip knowledge graph system SmarttripKG,which cover all kind of entities in tourism vertical domain,such as 110K sights,230K shopping center,400K hotels and nearly 1M restaurant.In a word,our SmarttripKG contains more than 10M tourism knowledge triples,which provides a tourism knowledge support for enhancing travel QA bot.