Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable CTR Prediction via LLM Reasoning

Xiaohan Yu, Li Zhang, Chong Chen

TL;DR

This work addresses the opacity of CTR models by introducing ExpCTR, an integrated framework where LLM-based explanations are generated and optimized within the CTR prediction loop. Using proximal policy optimization with two alignment rewards (LC Alignment and IC Alignment) and a lightweight LoRA-based three-stage training paradigm, ExpCTR trains explanations that are both user-intentful and consistent with the underlying CTR model. Empirical results across BookCrossing, MovieLens-20M, and Amazon Books show that ExpCTR-LLM improves explanation-quality-driven scoring over post-hoc baselines, while ExpCTR-Aug leverages textual explanations to boost CTR accuracy substantially (e.g., up to 18.2% AUC gains on certain datasets). The approach eliminates the need for large, curated explanation datasets and highlights the potential for RL-guided, language-model-based interpretability to enhance both trust and performance in recommender systems.

Abstract

Recommendation Systems have become integral to modern user experiences, but lack transparency in their decision-making processes. Existing explainable recommendation methods are hindered by reliance on a post-hoc paradigm, wherein explanation generators are trained independently of the underlying recommender models. This paradigm necessitates substantial human effort in data construction and raises concerns about explanation reliability. In this paper, we present ExpCTR, a novel framework that integrates large language model based explanation generation directly into the CTR prediction process. Inspired by recent advances in reinforcement learning, we employ two carefully designed reward mechanisms, LC alignment, which ensures explanations reflect user intentions, and IC alignment, which maintains consistency with traditional ID-based CTR models. Our approach incorporates an efficient training paradigm with LoRA and a three-stage iterative process. ExpCTR circumvents the need for extensive explanation datasets while fostering synergy between CTR prediction and explanation generation. Experimental results demonstrate that ExpCTR significantly enhances both recommendation accuracy and interpretability across three real-world datasets.

Explainable CTR Prediction via LLM Reasoning

TL;DR

This work addresses the opacity of CTR models by introducing ExpCTR, an integrated framework where LLM-based explanations are generated and optimized within the CTR prediction loop. Using proximal policy optimization with two alignment rewards (LC Alignment and IC Alignment) and a lightweight LoRA-based three-stage training paradigm, ExpCTR trains explanations that are both user-intentful and consistent with the underlying CTR model. Empirical results across BookCrossing, MovieLens-20M, and Amazon Books show that ExpCTR-LLM improves explanation-quality-driven scoring over post-hoc baselines, while ExpCTR-Aug leverages textual explanations to boost CTR accuracy substantially (e.g., up to 18.2% AUC gains on certain datasets). The approach eliminates the need for large, curated explanation datasets and highlights the potential for RL-guided, language-model-based interpretability to enhance both trust and performance in recommender systems.

Abstract

Recommendation Systems have become integral to modern user experiences, but lack transparency in their decision-making processes. Existing explainable recommendation methods are hindered by reliance on a post-hoc paradigm, wherein explanation generators are trained independently of the underlying recommender models. This paradigm necessitates substantial human effort in data construction and raises concerns about explanation reliability. In this paper, we present ExpCTR, a novel framework that integrates large language model based explanation generation directly into the CTR prediction process. Inspired by recent advances in reinforcement learning, we employ two carefully designed reward mechanisms, LC alignment, which ensures explanations reflect user intentions, and IC alignment, which maintains consistency with traditional ID-based CTR models. Our approach incorporates an efficient training paradigm with LoRA and a three-stage iterative process. ExpCTR circumvents the need for extensive explanation datasets while fostering synergy between CTR prediction and explanation generation. Experimental results demonstrate that ExpCTR significantly enhances both recommendation accuracy and interpretability across three real-world datasets.

Paper Structure

This paper contains 26 sections, 16 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminary
  4. Problem Definition
  5. Theoretical Basis of ExpCTR
  6. Methodology
  7. Explanation Generation
  8. Reward Design
  9. Proximal Policy Optimization
  10. Explanation and LLM-CTR Alignment (LC Alignment)
  11. Explanation and ID-CTR Alignment (IC Alignment)
  12. Training Paradigm
  13. Light Weight Tuning
  14. Iterative Training
  15. Experiment
  16. ...and 11 more sections

Figures (4)

  • Figure 1: Comparison of current post-hoc paradigm methods and ours.
  • Figure 2: The overall architecture of ExpCTR.
  • Figure 3: Trends of LC alignment and IC alignment rewards across iterative training on ML-20M and BookCrossing datasets.
  • Figure 4: Hyperparameter sensitivity of ExpCTR in KL penalty $\beta$ and reward normalization bound $\delta$.