[5Yin-A] Poster 8

[5Yin-A-02] Proposal on Refining Synthetic Data Based on User Feedback to Recommender Systems

〇Takumi Miyazawa¹, Hiroki Shibata¹, Yasufumi Takama¹ (1. Tokyo Metropolitan University)

Keywords:

Recommendation,Synthetic data,Data augmentation

This paper proposes a method to refine synthetic datasets for recommender systems using user feedback.Collaborative filtering, a key technique in recommender systems, recommends itemsbased on users' rating histories without requiring user or item features.However, collecting real histories can be challenging. As a solution, synthetic data, which is generated to mimicthe statistical properties of real data, has been studied in machine learning.In recommender systems, methods for generating synthetic data for collaborative filtering have also been proposed.However, modifying the characteristics of existing synthetic data often requires a regeneration of the entire dataset.To address this issue, this paper aims to improve data quality by refining existing synthetic data from the perspective of recommendation effectiveness.The proposed method treats the recommendation accuracy for each user as feedback to the recommender system,and adds synthetic data related to users with low recommendation accuracy for improving the recommendation quality to them.Evaluation experiments demonstrate the effectiveness of the proposed method in improving both recommendation accuracy and diversity.