Personalized Attraction Recommendation Algorithm Based On Knowledge Graph And Deep Learning

With the improvement of people's living standards,people's demand for tourism is increasing.Faced with a huge amount of Internet information,traditional search engines often cannot meet the personalized needs of users in terms of attraction recommendation,and users usually need to spend a lot of time filtering out attraction information.Personalized attraction recommendation is one of the important topics of smart tourism cities and location services.How to make accurate recommendations for users,provide personalized services,and improve user satisfaction is particularly important.For this reason,the personalized attraction recommendation system has attracted much attention and has become one of the hot spots in academia and business research.The knowledge graph effectively solves the problems of item cold start and data sparseness.At the same time,its rich graph network structure has caused many related scholars to study and apply it to recommendation systems.In recent years,knowledge graph based on recommendation algorithms have shown strong feasibility and robustness,but existing knowledge graph algorithms have a preference that cannot be modeled for users.Although sequence based on recommendation algorithms model user behavior sequences,they often do not consider item attribute information.In order to overcome these shortcomings,this paper proposes the KG-ULSP model and the KGSA4 Rec model to be applied to a personalized attraction recommendation system.The specific research contents of this article are as follows:(1)A personalized attractions recommendation method based on knowledge graph and user's long-term short-term and preferences(KG-ULSP)is proposed.First,construct a knowledge graph of users,attractions,and ratings of tourist attractions,use the representative method Node2 Vec in network representation learning,and use the neural network language model Word2 Vec to learn each a feature vector representations of attractions using paranoid random walk strategy.The attraction vector is then input to a gated recurrent unit(GRU)network training to obtain a latent vector representation of the sequence of attractions.Finally,the attention mechanism is used to model the user's long-term and short-term preferences,linearly stitching the user's long-term and short-term preferences,predicting the probability of each visit at the next visit,and generating a recommended list of attractions that users may like.The experimental results prove that the knowledge graph and user's long-term and short-term preference models are more accurate than other model recommendation results.It also shows that the attraction vector that combines the attributes of the attraction and the semantic information of the network structure,and the modeling of long-term and short-term preferences can improve the effectness of the recommendation results.(2)Considering the rich graph network structure of knowledge graphs,we further perfect the attribute information of attractions,and introduce a translation model method into a personalized attraction recommendation system,and propose a personalized attractions recommendation method based on knowledge graph and self-attention mechanism.The TransD translation model in knowledge representation learning is used to map the tourists attractions and the six attributes of the attractions(the tourist's rating of the attractions,the geographical location of the attractions,the type,the level(5A),the tickets,the seasons suitable for playing),and are mapped separately.In different implicit semantic spaces,learn the feature representation of attractions,then use the self-attention mechanism to model the sequence behavior,mine the preference information of tourists on different attractions,and finally predict the scores of tourists on the predicted attractions through matrix decomposition,and generate recommended list.The experiments on real travel data and a lot of comparative experiments verify the effectiveness of the proposed method.