Explicable recommendation model based on a time‐assisted knowledge graph and many‐objective optimization algorithm
Rui Zheng,Linjie Wu,Xingjuan Cai,Yubin Xu
DOI: https://doi.org/10.1002/cpe.8210
2024-06-26
Concurrency and Computation Practice and Experience
Abstract:Summary Existing research on recommender systems primarily focuses on improving a single objective, such as prediction accuracy, often ignoring other crucial aspects of recommendation performance such as temporal factor, user satisfaction, and acceptance. To solve this problem, we proposed an explicable recommendation model using many‐objective optimization and a time‐assisted knowledge graph, which utilizes user interaction times within the graph to prioritize recommending recently frequently visited items and is further optimized using a many‐objective optimization algorithm. In this model, the temporal weight of user actions at different times is first determined through a time decay function. Additionally, if a user clicks on the same item again, the current action's temporal weight is set to one. This strategy prioritizes recent user actions and frequently visited items, reflecting current interests and preferences better. Next, the created knowledge graph is used to create a list of potential recommendations. Embedding methods obtain the vectors for entities and relations in the path. These vectors, combined with the temporal weight of actions, quantify the explainability of user recommendations. Optimizing the rest of the recommendation performance with many objective algorithms while focusing on the user's recent frequent visits to the item. Finally, the outcomes of the research study indicate that, compared to other explicable recommended methods, our model, considering temporal factor, improved average accuracy by 11%, diversity by 1%, and explainability by 21% in the Useraction1 data set. Results in other data sets also indicate that the proposed model maintains accuracy, diversity, and novelty while enhancing explainability.
computer science, theory & methods, software engineering