Adaptive User Interest Modeling via Conditioned Denoising Diffusion For Click-Through Rate Prediction

2025-09-24

AI Summary

Existing user interest modeling approaches suffer from two key limitations: (1) treating behavioral sequences as noise-free preference signals, thereby ignoring pervasive observational noise; and (2) generating static, context-agnostic interest representations, failing to capture dynamic user intent. To address these, we propose the Context-Aware Denoising Diffusion Framework (CADF), the first to introduce conditional denoising diffusion into user interest modeling—enabling a paradigm shift from conventional “identify-and-aggregate” to controllable purification. CADF leverages query-user-item-context quaternary interaction features to guide both forward noise injection and conditional reverse denoising, while integrating category-aware behavioral sequence filtering for robust sequence purification. Consequently, it dynamically generates context-sensitive interest representations. Extensive offline evaluations and large-scale online A/B tests demonstrate that CADF consistently outperforms state-of-the-art methods, achieving significant improvements in click-through rate prediction performance.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Intelligent Query Processing
• Knowledge Representation and Reasoning: Preferences

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Content-based information diffusion
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

User behavior sequences in search systems resemble "interest fossils", capturing genuine intent yet eroded by exposure bias, category drift, and contextual noise. Current methods predominantly follow an "identify-aggregate" paradigm, assuming sequences immutably reflect user preferences while overlooking the organic entanglement of noise and genuine interest. Moreover, they output static, context-agnostic representations, failing to adapt to dynamic intent shifts under varying Query-User-Item-Context conditions. To resolve this dual challenge, we propose the Contextual Diffusion Purifier (CDP). By treating category-filtered behaviors as "contaminated observations", CDP employs a forward noising and conditional reverse denoising process guided by cross-interaction features (Query x User x Item x Context), controllably generating pure, context-aware interest representations that dynamically evolve with scenarios. Extensive offline/online experiments demonstrate the superiority of CDP over state-of-the-art methods.

Problem

• Modeling user interests from noisy behavior sequences with exposure bias
• Generating dynamic interest representations that adapt to contextual conditions
• Overcoming limitations of static aggregation methods for CTR prediction

Innovation

• Conditional denoising diffusion for user interest modeling
• Generating dynamic representations from cross-interaction features
• Purifying user behavior sequences contaminated by noise