Enhancing Time Awareness in Generative Recommendation

TL;DR

This work addresses the gap in generative recommender systems by integrating temporal dynamics into the generation and ranking process. GRUT introduces Time-aware Prompting with user-level temporal context and item-level transition context, combined via Context-integrated ID Generation, and enhances rankings through Trend-aware Inference that incorporates recent item trends without retraining. Empirical results across four real-world datasets demonstrate state-of-the-art or competitive performance, with notable gains for long time intervals and clear ablation evidence of each component’s contribution. The approach offers practical, low-overhead trend adjustments and highlights the importance of time awareness in LLM-based recommendation pipelines, enabling more timely and personalized suggestions.

Abstract

Generative recommendation has emerged as a promising paradigm that formulates the recommendations into a text-to-text generation task, harnessing the vast knowledge of large language models. However, existing studies focus on considering the sequential order of items and neglect to handle the temporal dynamics across items, which can imply evolving user preferences. To address this limitation, we propose a novel model, Generative Recommender Using Time awareness (GRUT), effectively capturing hidden user preferences via various temporal signals. We first introduce Time-aware Prompting, consisting of two key contexts. The user-level temporal context models personalized temporal patterns across timestamps and time intervals, while the item-level transition context provides transition patterns across users. We also devise Trend-aware Inference, a training-free method that enhances rankings by incorporating trend information about items with generation likelihood. Extensive experiments demonstrate that GRUT outperforms state-of-the-art models, with gains of up to 15.4% and 14.3% in Recall@5 and NDCG@5 across four benchmark datasets. The source code is available at https://github.com/skleee/GRUT.

Figures (7)

• Figure 1: Illustration of our motivation. While (a) existing generative recommenders only consider sequential order, (b) our method utilizes temporal information.
• Figure 2: Overall architecture of G RUT. The core innovation of the model is (a) Time-aware Prompting that captures evolving user preferences. This is followed by (b) Context-integrated ID Generation that aggregates contexts and complemented by (c) Trend-aware Inference that adaptively incorporates the trend of items.
• Figure 3: Performance comparison across time interval groups, defined by the number of days between each user’s most recent interaction and the target item.
• Figure 4: Performance of G RUT over varying the number of neighboring items $k$ in $C_v$.
• Figure 5: Performance of G RUT over varying the window size $N$ in the trend score.