Lightweight Modality Adaptation to Sequential Recommendation via Correlation Supervision
Dec. 15th, 2023: Got accepted by ECIR ‘24. Congrats to Hengchang!
Hengchang HU, Qijiong LIU, Chuang LI, Min-Yen KAN#
[Code] [Paper]
Abstract
In Sequential Recommenders (SR), encoding and utilizing modalities in an end-to-end manner is costly in terms of modality encoder sizes. Two-stage approaches can mitigate such concerns, but they suffer from poor performance due to modality forgetting, where the sequential objective overshadows modality representation. We propose a lightweight knowledge distillation solution that preserves both merits: retaining modality information and maintaining high efficiency. Specifically, we introduce a novel method that enhances the learning of embeddings in SR through the supervision of modality correlations. The supervision signals are distilled from the original modality representations, including both (1) holistic correlations, which quantify their overall associations, and (2) dissected correlation types, which refine their relationship facets (honing in on specific aspects like color or shape consistency). For the issue of rapidly overshadowed modal information, we propose an asynchronous learning step, allowing the original information to be preserved longer for training the representation learning module. Our approach is compatible with various backbone architectures, and it outperforms the top baselines by 6.8% on average. We empirically demonstrate that preserving original feature associations from modality encoders significantly aids in the recommendation task-specific adaptation. Additionally, we find that larger modality encoders (e.g., Large Language Models) contain richer feature sets and also necessitate more fine-grained modeling to reach their full performance potential.
Citation
1 | @inproceedings{hu2024lightweight, |
Lightweight Modality Adaptation to Sequential Recommendation via Correlation Supervision