| In recent years,"information overload" has become one of the major problems faced by big data on the Internet.In order to solve the problem of "information overload",the recommendation system mines the interest of users from the massive amount of data and implemented personalized recommendation services.The traditional collaborative filtering methods mainly utilize the interaction information or attributes information of users and items to make recommendations.When the interaction data is sparse,the accuracy of recommendation is low.The recommendation results are difficult to reach the satisfaction of users.Different types of similar neighbors can provide rich collaborative semantic interest to the target user and help improve the performance of the recommendation system.Existed approaches ignore the roles of multi-scale attribute neighbors and multi-scale interaction neighbors in modeling the user’s semantic interest representation.In this paper,we propose an attribute neighbor attention-aware rating prediction recommendation method and a rating prediction recommendation method by fusing the attention of multi-scale attribute neighbors and interaction neighbors,and validate the effectiveness of the method through experiments.The main research content are as follows:(1)A multi-scale attribute neighbor attention-aware rating prediction recommendation method is proposed.First,different scales of user attribute neighbor views are constructed based on the similarity of their attribute values,and the embedding representations of user nodes on different attribute neighbor views are modeled by graph neural network representation learning.Then,the semantic importance of node attribute on different attribute neighbor views is learnt with an attention mechanism,and the multi-scale neighbor view attribute semantic representation of nodes is fused.Finally,the rating prediction is performed through the inner product of users and items.Experimental results on the Movielens-100 kr and Book Crossing-10 Core datasets show that the collaboration of attribute neighbors at different scales can enhance the semantic representation of nodes and improve the accuracy of the recommender system.(2)A rating prediction recommendation method by incorporating multi-scale attribute neighbors and interaction neighbors’ attention is proposed.First,based on the user-item interaction matrix,different scales of interaction neighbor views are constructed based on the user’s rating behavior preferences,and different types of neighbors are obtained by combining the multi-scale user attribute neighbor views.Further,the node embedding representations of users and items under the attribute view and the node embedding representations under the interaction view are modeled by graph neural networks.Then,the two representations of nodes are fused with an attention mechanism and the rating prediction is performed.Finally,the validity of the model is verified through experiments.The experimental results show that the fusion of multi-scale attribute neighbors and interaction neighbors can improve the performance of the recommendation system and can alleviate the problem of poor recommendation results under the sparse scenario of user-item interaction matrix.(3)A collaborative multi-scale neighbor recommendation system for tourist spots is designed and implemented.The system makes use of the attribute information of users and tourist spots as well as the interaction information of user for tourist spots to get the neighbor information on different scales of neighbor views,and models the representation of users and tourist spots,and provides users with personalized recommendations for tourist spots and other functions. |
