| Recommendation systems are crucial in the age of the digital economy,as is obvious.Despite the fact that recommender systems have become a core function of many ecommerce businesses,there are still some obstacles to overcome.The most serious issues are data sparsity and the cold start problem.Researchers have raised the issue of improving recommendation systems by combining social relationships with the development of social networks.Based on the theory of social homogeneity,most social recommendation algorithms believe that the preferences of friends are similar and only focus on the explicit relationship of users,i.e.,the users connected to them in the social network,to infer the preferences of users themselves.However,the explicit relationship only utilizes the trust relationship,and a single trust relationship has a lot of noise data.For example,users and their friends do not necessarily have similar interests,and directly using the explicit relationship is not conducive to mining reliable user preferences.Furthermore,the accumulation of explicit social relations takes time,so there is still the issue of sparse data,resulting in a very limited recommendation effect.In light of these issues,the following research was conducted:(1)Chapter 3 proposes a recommendation framework that fuses explicit-implicit relational hypergraph networks and contrast learning.This recommendation model mines potential connections between users from strangers who have similar interests to them,which are called implicit relationships,and then fuses them with explicit relationships for recommendation.The framework is divided into two parts.The first is the main task of recommendation based on hypergraph network,which uses hypergraph to build explicit and implicit relationship graphs and hypergraph convolutional neural network to capture highorder relationships between users and form reliable user embedding.The other part is the auxiliary task of self-supervised comparative learning,which increases the supervised data through comparative learning to alleviate the recommendation system’s problem of sparse user-item interaction data.Experiments on two public data sets were carried out to validate the model’s validity.Experimental results show that this model can perform well on data sets with high social density,and self-supervised comparative learning can improve recommendation performance.(2)The social recommendation model with explicit and implicit relationships proposed in Chapter 3 does not take into account the impact of data sets with varying social densities on the model,as well as the high complexity of contrast learning due to the use of structural disturbance in the model.As a result,we propose a social recommendation framework based on hypergraph contrast learning without graph enhancement.The framework is divided into two parts.One method is to encode user embedding from various perspectives using explicit and implicit diagrams.Simultaneously,taking into account the contribution of various subgraphs to recommendation,it employs the aggregation of selfattention mechanism to generate comprehensive user representation,thereby addressing the impact of data set on the model.The other part is an auxiliary task of self-supervised contrast learning with Graph-Augmentation-Free,which can alleviate data sparsity by introducing uniform noise into the embedded space for contrast learning.Experiments were performed on three public data sets to validate the model’s validity.The experimental results show that the model can ensure recommendation reliability while also ensuring more efficient operation,as well as alleviating the cold start problem and long tail phenomenon.(3)This paper designs and implements a movie recommendation system based on requirements analysis,introduces the overall framework and functional design steps of the system in detail,and then displays various functions supported by this platform in a graphical manner in order to more vividly verify the effectiveness of the model in real scenes. |
