At the 30th ACM Conference on Knowledge Discovery and Data Mining (KDD 2024) held in Barcelona, Spain, from August 25th to 29th, a team led by Prof. CHEN Enhong from the University of Science and Technology of China (USTC), in collaboration with Huawei, won the Best Student Paper Award for their work titled Dataset Regeneration for Sequential Recommendation.
Prof. CHEN’s team won Best Student Paper Award at KDD 2024. (Image by USTC)
The paper addresses the challenges in sequential recommendation, a critical aspect of artificial intelligence that focuses on next-token prediction. Existing approaches of sequential recommendation primarily follow a model-centric paradigm, concentrating on designing complex models and training strategies while often overlooking the quality of the data. The USTC team, for the first time, introduced a novel data-centric approach that emphasizes on how to obtain high-quality, information-rich training datasets to ultimately enhance model performance.
The framework they proposed, DR4SR, represents a data-centric paradigm that regenerates training datasets to improve generalizability across different model architectures. The framework pretrains diverse data regenerators and utilizes hybrid inference method to transform the original sequential dataset into a more training-friendly dataset, which allows various baseline prediction models to achieve enhanced recommendation performance. The proposed methodology was tested across four widely-used datasets, all resulting in significant performance improvements.
Additionally, the team developed DR4SR+, a model-aware dataset personalization framework, which fine-tunes the regenerated dataset for specific target models.
ACM KDD is considered one of the most prestigious conferences in data mining, with rigorous requirements for the innovation, technical advancement, system completeness, and writing quality of research papers. KDD 2024 received over 2,000 submissions, and only one paper with student as the first author was selected as the Best Student Paper. This marks the second time that Prof CHEN Enhong’s team has won Best Student Paper Award at KDD.
The team’s innovative research not only boosts the performance of existing recommendation models but also marks a shift towards a data-centric paradigm in sequential recommendation.
Paper link: https://dl.acm.org/doi/10.1145/3637528.3671841
(Written by CHEN Yehong, edited by ZHANG Yihang, USTC News Center)
