Research on Optimization of Recommendation Model Based on Value Evaluation: A Case Study of Amazon Dataset

Abstract

The core objective of a recommendation system is to precisely match items that users might be interested in. Among them, how to provide personalized recommendations for different user groups has become a key research issue. Currently, mainstream recommendations adopt a single-model approach, and the recommendation results are usually only applicable to a single type of user group. This paper aims to combine the evaluation and reasoning capabilities of large models, adopt an innovative recommendation process, and use different recommendation algorithms for different user groups to obtain a series of recommendation results. By leveraging the comprehensive evaluation capabilities of the Deepseek-V3.1 large model, the optimal recommendation small models for different user groups are provided. Finally, the reasoning capabilities of the large model are utilized to give the parameter optimization scheme for the model, achieving the optimization of the model effect. This paper conducted experimental verification using the Amazon dataset, and used models such as the heuristic recall algorithm and the dual-tower model algorithm for model recommendations. Through the above process, model screening and optimization were carried out, and the optimized results were obtained. After comparison, it was found that the model's accuracy rate increased by 34.88% and the recall rate increased by 40.74%, verifying the hypothesis of the experiment.