Research Of Dining Recommendation Method Based On Knowledge Graph

With the improvement of the quality of human life,the catering industry has gradually become a key area of the market economy.Faced with a variety of restaurants,consumers don’t know how to choose.The recommendation algorithm can extract consumers’ subconscious needs without clear needs,and provide consumers with good choice suggestions.However,some traditional recommendation algorithms are often troubled by factors such as data sparsity and cold start.Knowledge graph contains a wealth of knowledge,treating it as a source of auxiliary information into the recommendation task can improve diversity,accuracy,and interpretability.Knowledge graph has a lot of missing knowledge.However,the existing recommendation methods based on knowledge graph generally believe knowledge graph is complete,and only transmits knowledge at the shallow level of the original entity data,which will limit the recommendation performance.And common knowledge graph recommendation methods generally only use the entity information,and rarely use relation information.In order to make full use of the information provided by the knowledge graph,and consider the incompleteness of the knowledge graph,and better understand consumers’ preferences for a certain restaurant,DRKG model is proposed.The new model uses a multi-task learning method to simultaneously perform recommendation task and knowledge graph embedding task.DRKG model uses Cross&Compress Module and KG Relation&Preference Joint Module to connect the two tasks and learns from each other to make up each missing information.To make the recommendation method more interpretable,DRKG uses KG Relation&Preference Joint Module to integrate the relation embeddings into the recommendation algorithm while the knowledge graph transfers knowledge to understand a consumer’s preference in a fine-grained manner,and deeply understand the reasons why the consumer likes a certain restaurant.Experiments verified the knowledge graph plays a positive role in the recommendation algorithm.It proved DRKG has advantages in understanding user’s preference and has certain practical value in the dining recommendation scenario.Compared with other recommendation algorithms based on knowledge graph,DRKG performs well in both ClickThrough Rate(CTR)prediction and Top-K prediction.In the CTR prediction,the AUC index reached 87.5%,and the ACC index reached 79.6%.