Research On Travel Recommendation Algorithm Based On Knowledge Graph And User Dynamic Preference

There are challenges in travel recommendation for more fine-grained description of user preferences and complex information of travel products,as well as the problems of sparse user data and variable interests in online travel information recommendation.And in the field of Chinese knowledge graph,there is no publicly available knowledge graph of tourist attractions.However,unlike previous recommendations,there is a certain scarcity of tourism recommendation data,both user interaction data and attraction attribute information,which affects the effectiveness of recommendations.Knowledge graph-based tourism recommendation can take advantage of its own structure and rich information as an aid to the recommendation system to provide data.In order to solve the above problems,the main research of this paper is as follows:(1)The construction of knowledge graph in tourism field,and the construction method of tourism knowledge graph is designed and implemented.The concept of knowledge graph is introduced into the tourism recommendation domain,the knowledge features of the tourism domain are analyzed,the division of entities and relationships in the domain is realized,and the construction of the tourism domain ontology library is completed.Then the entities and relations in the tourism domain are extracted,and the knowledge graph storage based on the relational database is completed.Finally,the pros and cons of relational database for knowledge storage are analyzed,and the knowledge storage method is improved by using Neo4 j.(2)To address the problems of sparse data and variable user interests in online travel information recommendation,this study proposes an online travel information recommendation model(Knowledge Graph Embedding Travel Recommendation,KGETR).The model includes two core modules of information embedding and interest propagation.The information embedding module uses knowledge graph embedding to provide a unified vector representation of tourism information and user attributes;the interest propagation module enriches the attribute features of users and tourism products through interest propagation based on the combination of embedding and path.In addition,this paper also introduces temporal attributes to construct knowledge graph quadruples to accurately grasp users’ interests.The results show that the accuracy rate,recall rate and F1 are improved by 6.72%,13.8% and 29.03%,respectively.(3)To address the problems of complex long-term preferences of users and more fine-grained description of complex information of travel products,a neural attention travel recommendation model(Long-Short Attention Travel Recommendation,LSATR)that introduces knowledge graph representation into the recommendation system is constructed by combining long-term and short-term preferences of users.Specifically,LSATR is divided into two main core modules,namely,the travel product encoder and the user encoder.The travel product encoder is a multi-channel,word-entity aligned knowledge-aware convolutional neural network that incorporates semantic-level and knowledge-level representations of travel products and treats words and entities as multiple channels,explicitly maintaining their alignment relationships during the convolution process.The user encoder uses a bidirectional long-and short-term memory neural network(Bi-directional Long-Short Term Memory,Bi-LSTM)to study users’ long-and short-term preferences.To satisfy the different interests of users,an attention module is designed to dynamically aggregate users’ histories about current travel product candidates.Extensive experiments on a real travel e-commerce dataset demonstrate that LSATR achieves substantial improvements over existing deep recommendation models.