| The intelligent development of the Internet has promoted the rapid development of catering ecommerce websites and software applications,and the recommendation system has been widely used in catering service websites.With the development of recommendation systems,users pay more and more attention to the personalization and interpretability of recommendation results.However,most existing restaurant recommendation methods consider recommendation similarity only from the user’s perspective(e.g.,reviews,sentiment,ratings,etc.),ignoring the attribute information of the restaurant and the higher-order relationships between restaurants.Some traditional recommendation methods use only users’ explicit or implicit feedback on restaurants to make recommendations,which is prone to the problem of data sparsity.Besides,the dining interests of users are diverse and not fixed.Different users have different needs and preferences for the style,environment and facilities of the restaurant.In order to obtain the recommendation results more in line with users’ needs,the dynamic changes of users’ interests should be fully considered.In order to solve the above problems,the main research work of this paper is as follows:(1)In this paper,we introduce knowledge graphs as auxiliary information to link the interaction matrix between users and restaurants.By analyzing the semantic structure of the knowledge graph triplet,the higher-order connections and semantic information between the user preferred restaurant and the candidate restaurant,and between the restaurant and the restaurant are mined to improve the accuracy and interpretability of the recommendation.(2)In this paper,based on real dining data from Yelp website,we filter and analyze restaurant attributes,classify attribute categories,create graphical relationships,design and build knowledge graph model layer and data layerand to construct a restaurant knowledge graph R-KG for analysis and research.Using the mainstream graph database Neo4 j for storage,we build a visual restaurant knowledge graph database that can be queried and analyzed.(3)Aiming at the problem of poor performance of existing recommendation algorithms under dynamic changes of user interests and sparse data scenarios,this paper proposes a restanrant recommendation method that combines knowledge graphs and users’ long-term and short-term interests(RR-KGLSI).Fully considering the impact of user interest changes on user dining feature extraction,this paper learns user dining preferences from both long-term stable interest and shortterm dynamic interest,and models long-term and short-term interest using knowledge graph semantic structure and neural network respectively.Considering the degree of distinguishing user preferences for each attribute of the restaurant,this paper classifies different attributes and calculates the correlation degree for different categories of attributes separately through the semantic structure of the knowledge graph to distinguish user preferences.(4)Considering the effect of long and short-term interest fusion on the recommendation results,this paper designs a gating structure to dynamically obtain the weight of interest assignment,i.e.,the ratio of long and short-term interest contribution.This structure can adaptively fuse the long-term and short-term interests of users to generate mixed interest representation according to different user interests.In addition,this paper compares this interest fusion method with common connection fusion methods.The performance indexes of AUC,ACC and F1 are 0.9310,0.8724 and 0.8773 respectively,which are better than those of MLP,FM,Deep FM and Ripple Net. |
