Dataset Regeneration for Sequential Recommendation

TL;DR

This work proposes a novel data-centric paradigm for developing an ideal training dataset using a model-agnostic dataset regeneration framework called DR4SR, and introduces the DR4SR+ framework, which incorporates a model-aware dataset personalizer to tailor the regenerated dataset specifically for a target model.

Abstract

The sequential recommender (SR) system is a crucial component of modern recommender systems, as it aims to capture the evolving preferences of users. Significant efforts have been made to enhance the capabilities of SR systems. These methods typically follow the model-centric paradigm, which involves developing effective models based on fixed datasets. However, this approach often overlooks potential quality issues and flaws inherent in the data. Driven by the potential of data-centric AI, we propose a novel data-centric paradigm for developing an ideal training dataset using a model-agnostic dataset regeneration framework called DR4SR. This framework enables the regeneration of a dataset with exceptional cross-architecture generalizability. Additionally, we introduce the DR4SR+ framework, which incorporates a model-aware dataset personalizer to tailor the regenerated dataset specifically for a target model. To demonstrate the effectiveness of the data-centric paradigm, we integrate our framework with various model-centric methods and observe significant performance improvements across four widely adopted datasets. Furthermore, we conduct in-depth analyses to explore the potential of the data-centric paradigm and provide valuable insights. The code can be found at https://github.com/USTC-StarTeam/DR4SR.

Figures (7)

• Figure 1: Model-centric paradigm v.s. Data-centric pradigm.
• Figure 2: The framework of the proposed data-centric paradigm: (A) The Pre-training stage of DR4SR involves training a diversity-promoted data regenerator utilizing the curated pre-training dataset. (B) The inference stage of DR4SR regenerates each source sequence into multiple target patterns with a hybrid inference strategy. (C) Model-aware dataset regeneration with a personalizer further tailors the regenerated dataset to each target model.
• Figure 3: Pre-training task construction. Assuming two given sequences (1,2,3,4,5), (1,2,3) with a window size 3 and a threshold 2, the following patterns can be extracted: (1,2), (1,3), (2,3), (1,2,3). This is because these patterns appear twice within the sliding window. Then the regenerator is supposed to regenerate each sequence into multiple corresponding patterns.
• Figure 4: Scores assigned by the dataset personalizer for different target models on Toys. From left to right, the variances of the scores for each target model are 0.0280, 0.0244, 0.0241, 0.0248, and 0.0247.
• Figure 5: Relative NDCG@20 improvement of graphs and data augmentations on different datasets.