Research On Personalized Recommendation Algorithm Based On Fine-grained Intention

With the rapid development of Internet technologies,videos,news,and other information in the top media show explosive growth.How to find related information quickly that also matches users’ needs precisely has become a fashionable topic in recommendation systems.Personalized recommendation algorithms analyze the historical interaction data to find the users’ potential intentions.It intensifies user-item relations,filters out sophisticated information,and improves the users’ experience and participation.However,it hard to distinguish embeds features and interactive relations.There are still some problems such as flat structure of intentional representation.To address those problems,this paper is based on fine-grained intent for the user in-depth research into the influence of idiosyncratic behavior.The main research work of this paper is reproduced below.(1)This paper introduces a Hierarchical Intent Disentangling for Graph Convolution Neural Network Recommendation model which divides the user-item interaction graph into multiple dynamic interaction sub-graphs to show the hierarchy of user intentions from fine-grained to coarse-grained.First,this model adaptively fuses information from highorder neighborhoods by its connectivity in each intent interaction sub-graph and disentangles the fine-grained intent representations of users from it.Second,it builds a intentional hierarchical structure from fine-grained to coarse-grained intentions by constructing coarse-grained intention nodes in the high-level network through the similarity between the fine-grained intentions in low-level relationships.Finally,it aggregates hierarchical intentions to get the representation of the combination of users and projects and does the inner product prediction and iterative optimization on them.(2)This paper presents a Sequential Recommendation Model Fusing Users’ Long-term and Short-term Intentions.It considers the users’ long-term and short-term intention as a probable factor in the fine-grained interactive behavior to deduct the real intention of users at a certain moment.Meanwhile,this paper designs a gated bidirectional recurrent network to solve the sequential dependency problem in the long-term sequences.It also employs an intentional change capture unit based on the attention mechanism to capture the change of users’ long-term intention.Moreover,this paper optimizes online meta-learning update strategy to model users’ short-term intentions in real time and the current timing information and content relevance into account.Finally,those two kinds of intentions are adaptively fused to predict the interaction probability at the next moment.In conclusion,this paper does the research on the personalized recommendation algorithms based on the fine-grained intention perception and proposes two models in this field which is the Hierarchical Intent Disentangling for Graph Convolution Neural Network Recommendation model and the Sequential Recommendation Model fusing users’ Long-term and Short-term Intentions respectively.Experiments and ablation studies have been done on those models to evaluate its performance and functionality.All those results demonstrate that the proposed methods can deduct the relations between users and projects by digging into the data of fine-grained intentions.Meanwhile,it can produce a more precise recommendation to meet the user’s diversity and personal requirements.