Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Effectiveness of LLMs in Temporal User Profiling for Recommendation

Milad Sabouri, Masoud Mansoury, Kun Lin, Bamshad Mobasher

TL;DR

This work investigates leveraging Large Language Models (LLMs) to model temporal dynamics in user preferences for recommender systems. It generates dual natural language profiles—short-term and long-term—from user histories, encodes them into semantic embeddings, and fuses them with an attention mechanism to obtain an interpretable user representation used for recommendations. Across Movies&TV and Video Games, the approach yields notable gains in more dynamic domains but shows limited benefits in sparser domains, highlighting a domain-dependent trade-off with computational costs. The study also demonstrates intrinsic interpretability via the natural-language profiles and attention weights, pointing to a path toward adaptive and transparent recommendations.

Abstract

Effectively modeling the dynamic nature of user preferences is crucial for enhancing recommendation accuracy and fostering transparency in recommender systems. Traditional user profiling often overlooks the distinction between transitory short-term interests and stable long-term preferences. This paper examines the capability of leveraging Large Language Models (LLMs) to capture these temporal dynamics, generating richer user representations through distinct short-term and long-term textual summaries of interaction histories. Our observations suggest that while LLMs tend to improve recommendation quality in domains with more active user engagement, their benefits appear less pronounced in sparser environments. This disparity likely stems from the varying distinguishability of short-term and long-term preferences across domains; the approach shows greater utility where these temporal interests are more clearly separable (e.g., Movies\&TV) compared to domains with more stable user profiles (e.g., Video Games). This highlights a critical trade-off between enhanced performance and computational costs, suggesting context-dependent LLM application. Beyond predictive capability, this LLM-driven approach inherently provides an intrinsic potential for interpretability through its natural language profiles and attention weights. This work contributes insights into the practical capability and inherent interpretability of LLM-driven temporal user profiling, outlining new research directions for developing adaptive and transparent recommender systems.

Effectiveness of LLMs in Temporal User Profiling for Recommendation

TL;DR

This work investigates leveraging Large Language Models (LLMs) to model temporal dynamics in user preferences for recommender systems. It generates dual natural language profiles—short-term and long-term—from user histories, encodes them into semantic embeddings, and fuses them with an attention mechanism to obtain an interpretable user representation used for recommendations. Across Movies&TV and Video Games, the approach yields notable gains in more dynamic domains but shows limited benefits in sparser domains, highlighting a domain-dependent trade-off with computational costs. The study also demonstrates intrinsic interpretability via the natural-language profiles and attention weights, pointing to a path toward adaptive and transparent recommendations.

Abstract

Effectively modeling the dynamic nature of user preferences is crucial for enhancing recommendation accuracy and fostering transparency in recommender systems. Traditional user profiling often overlooks the distinction between transitory short-term interests and stable long-term preferences. This paper examines the capability of leveraging Large Language Models (LLMs) to capture these temporal dynamics, generating richer user representations through distinct short-term and long-term textual summaries of interaction histories. Our observations suggest that while LLMs tend to improve recommendation quality in domains with more active user engagement, their benefits appear less pronounced in sparser environments. This disparity likely stems from the varying distinguishability of short-term and long-term preferences across domains; the approach shows greater utility where these temporal interests are more clearly separable (e.g., Movies\&TV) compared to domains with more stable user profiles (e.g., Video Games). This highlights a critical trade-off between enhanced performance and computational costs, suggesting context-dependent LLM application. Beyond predictive capability, this LLM-driven approach inherently provides an intrinsic potential for interpretability through its natural language profiles and attention weights. This work contributes insights into the practical capability and inherent interpretability of LLM-driven temporal user profiling, outlining new research directions for developing adaptive and transparent recommender systems.

Paper Structure

This paper contains 7 sections, 10 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Experiments
  5. Results and Discussion
  6. Ablation Study
  7. Conclusion

Figures (4)

  • Figure 1: Pipeline for LLM-Driven Temporal User Profile Generation and Recommendation.
  • Figure 2: A conceptual illustration of the framework’s potential for enhancing transparency. (A) and (B) show the short-term and long-term textual profiles generated and encoded by our current approach. (C) shows a hypothetical extension wherein the final recommendation is explicitly justified by both sets of preferences, further strengthening user-facing transparency—an aspect we plan to explore in future work.
  • Figure 3: Examples of Prompts for LLM-based Temporal User Profile Generation (Movies&TV Domain)
  • Figure 4: Relative gains in Recall@20 and NDCG@20 of the full model over ablation variants on the Movies&TV dataset.