Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Can We Predict the Next Question? A Collaborative Filtering Approach to Modeling User Behavior

Bokang Fu, Jiahao Wang, Xiaojing Liu, Yuli Liu

TL;DR

Problem: Static modeling of user preferences in LLM-driven dialogues limits proactive questioning. Approach: CFQP integrates a dynamic per-user memory with a user-user similarity graph to propagate preferences and guide next-question prediction, with a prediction step $\hat{q}_{t+1} = \text{LLM}(M_u,N_u,q_t)$. Contributions: (i) a novel CFQP framework combining memory and collaborative filtering, (ii) dynamic memory that evolves with interactions, (iii) collaborative reasoning and an error-correction loop, and (iv) strong empirical results on LexRAG and CrossWOZ showing improvements over standalone LLM baselines and ablations. Significance: advances personalized large language models toward proactive, anticipatory conversational agents that can guide dialogues more intelligently.

Abstract

In recent years, large language models (LLMs) have excelled in language understanding and generation, powering advanced dialogue and recommendation systems. However, a significant limitation persists: these systems often model user preferences statically, failing to capture the dynamic and sequential nature of interactive behaviors. The sequence of a user's historical questions provides a rich, implicit signal of evolving interests and cognitive patterns, yet leveraging this temporal data for predictive tasks remains challenging due to the inherent disconnect between language modeling and behavioral sequence modeling. To bridge this gap, we propose a Collaborative Filtering-enhanced Question Prediction (CFQP) framework. CFQP dynamically models evolving user-question interactions by integrating personalized memory modules with graph-based preference propagation. This dual mechanism allows the system to adaptively learn from user-specific histories while refining predictions through collaborative signals from similar users. Experimental results demonstrate that our approach effectively generates agents that mimic real-user questioning patterns, highlighting its potential for building proactive and adaptive dialogue systems.

Can We Predict the Next Question? A Collaborative Filtering Approach to Modeling User Behavior

TL;DR

Problem: Static modeling of user preferences in LLM-driven dialogues limits proactive questioning. Approach: CFQP integrates a dynamic per-user memory with a user-user similarity graph to propagate preferences and guide next-question prediction, with a prediction step . Contributions: (i) a novel CFQP framework combining memory and collaborative filtering, (ii) dynamic memory that evolves with interactions, (iii) collaborative reasoning and an error-correction loop, and (iv) strong empirical results on LexRAG and CrossWOZ showing improvements over standalone LLM baselines and ablations. Significance: advances personalized large language models toward proactive, anticipatory conversational agents that can guide dialogues more intelligently.

Abstract

In recent years, large language models (LLMs) have excelled in language understanding and generation, powering advanced dialogue and recommendation systems. However, a significant limitation persists: these systems often model user preferences statically, failing to capture the dynamic and sequential nature of interactive behaviors. The sequence of a user's historical questions provides a rich, implicit signal of evolving interests and cognitive patterns, yet leveraging this temporal data for predictive tasks remains challenging due to the inherent disconnect between language modeling and behavioral sequence modeling. To bridge this gap, we propose a Collaborative Filtering-enhanced Question Prediction (CFQP) framework. CFQP dynamically models evolving user-question interactions by integrating personalized memory modules with graph-based preference propagation. This dual mechanism allows the system to adaptively learn from user-specific histories while refining predictions through collaborative signals from similar users. Experimental results demonstrate that our approach effectively generates agents that mimic real-user questioning patterns, highlighting its potential for building proactive and adaptive dialogue systems.

Paper Structure

This paper contains 14 sections, 3 equations, 1 figure, 3 tables.

Table of Contents

  1. 1 Introduction
  2. 2 Related Work
  3. 3 Proposed Model:CFQP
  4. 3.1 Overall Framework
  5. 3.2 User Preference Learning and Collaborative Signal Mining
  6. 3.3 Dynamic Preference Modeling
  7. 3.4 Collaborative Reasoning and Error Correction
  8. 4 Experiments
  9. 4.1 Experimental Setup
  10. 4.2 Baselines
  11. 4.3 Performance Comparison with Standalone Large Models
  12. 4.4 Comparison of Different Backbone Models within the CFQP Framework
  13. 4.5 Ablation Study
  14. 5 Conclusion

Figures (1)

  • Figure 1: The overall architecture of the CFQP framework.The whole process is coordinated by three core modules: Learning, Modeling and Reasoning.