GOD: Enhancing Generalization via Deep Grafting for Sequential Recommendation
2026-08-17 • Information Retrieval
Information RetrievalMachine Learning
AI summaryⓘ
The authors address problems in sequential recommendation systems that happen when user history data is limited or noisy, which makes it hard to predict new user preferences. They propose a new method called Graft-Oriented Distillation (GOD), where parts of a teacher model and a student model are combined in different ways to better understand which parts of the student need improvement. This approach helps the student model generalize better without extra cost during use. Tests on real datasets show that GOD improves recommendation performance compared to existing methods.
Sequential recommendationKnowledge distillationTeacher-student modelsGraftingModel generalizationEmbeddingsEncoderOverfittingSparse dataHybrid models
Authors
WooJoo Kim, JunYoung Kim, JaeHyung Lim, HwanJo Yu
Abstract
Sequential recommenders often struggle with sparse and noisy histories, limiting generalization to unseen interactions. Knowledge distillation mitigates this by transferring dense supervision from a teacher to a student. However, most distillation methods run teacher and student independently, then match student outputs or representations to the teacher. Such supervision entangles student-component effects, blurring whether weak generalization stems from unreliable embeddings, overfitted encoding, or co-adaptation to sparse histories. In this paper, we propose Graft-Oriented Distillation (GOD), a component-level distillation framework for improved generalization through grafting. Grafting denotes replacing selected frozen-teacher components with trainable student counterparts to build hybrid source models. GOD uses these hybrid models to evaluate student embeddings with the teacher encoder and the student encoder with teacher embeddings, providing component-level feedback. At inference, GOD uses only the student, incurring no additional cost. Across three real-world datasets, GOD outperforms state-of-the-art baselines by up to 13.92%.