arXiv:2602.18907v2 Announce Type: replace 
Abstract: We introduce DeepInterestGR, a novel framework that integrates deep interest mining into the generative recommendation pipeline. This addresses the "Shallow Interest" problem - existing generative methods rely on surface-level textual features and fail to capture latent user motivations, limiting personalization depth and recommendation interpretability. Our approach leverages Multi-LLM Interest Mining (MLIM) via structured reasoning prompting, Reward-Labeled Deep Interest (RLDI) for quality control, and Interest-Enhanced Item Discretization (IEID) via RQ-VAE, combined with a two-stage SFT-GRPO training pipeline guided by an Interest-Aware Reward. We validate DeepInterestGR on three Amazon Review benchmarks (Beauty, Sports, Instruments), comparing against 14 state-of-the-art baselines including SASRec, BERT4Rec, TIGER, LC-Rec, and S-DPO. Our method achieves 5.8%-8.3% relative improvements on HR@10 and 7.7%-9.9% on NDCG@10 over the strongest baseline, with cross-domain generalization gains of +24.8%. These results provide evidence that incorporating deep semantic interests can effectively improve SID-based generative recommendation.

تم تقديم DeepInterestGR كإطار جديد يعزز أنظمة التوصية التوليدية من خلال معالجة قيود الأساليب الحالية، وخاصة مشكلة 'الاهتمام السطحي'. يدمج هذا الإطار استخراج الاهتمامات العميقة في خط أنابيب التوصية، مستخدمًا تعدين الاهتمامات متعدد النماذج اللغوية وعمليات تدريب على مرحلتين لتحسين التخصيص وقابلية التفسير.

DeepInterestGR se ha presentado como un marco novedoso que mejora los sistemas de recomendación generativa al abordar las limitaciones de los métodos existentes, especialmente el problema del 'interés superficial'. Este marco integra la minería de intereses profundos en el pipeline de recomendación, utilizando la Minería de Intereses Multi-LLM y un pipeline de entrenamiento en dos etapas para mejorar la personalización y la interpretabilidad.

DeepInterestGR a été introduit comme un cadre novateur qui améliore les systèmes de recommandation générative en s'attaquant aux limites des méthodes existantes, en particulier le problème de l'« intérêt superficiel ». Ce cadre intègre l'extraction d'intérêts profonds dans le pipeline de recommandation, utilisant le Multi-LLM Interest Mining et un pipeline d'entraînement en deux étapes pour améliorer la personnalisation et l'interprétabilité.

DeepInterestGR has been introduced as a novel framework that enhances generative recommendation systems by addressing the limitations of existing methods, particularly the 'Shallow Interest' problem. This framework integrates deep interest mining into the recommendation pipeline, utilizing Multi-LLM Interest Mining and a two-stage training pipeline to improve personalization and interpretability.