GenCI: Generative Modeling of User Interest Shift via Cohort-based Intent Learning for CTR Prediction

TL;DR

GenCI targets two core gaps in CTR prediction: rapid shifts in user intent and the disconnect between recall and ranking. It introduces a generative next item prediction objective to produce semantic interest cohorts via hierarchical quantization and a Transformer based NTP module, coupled with a hierarchical candidate aware network that refines and injects short term intent into the ranking stage. The framework endows CTR prediction with recall aware context and multi facet user intents by joint optimization of L CTR, L NTP and a self supervised regularizer, and demonstrates strong improvements on three public datasets with ablations supporting the value of each component. The work advances practical CTR systems by improving recall ranking alignment and robustness to evolving user interests, with implications for more effective and stable recommendations in dynamic environments.

Abstract

Click-through rate (CTR) prediction plays a pivotal role in online advertising and recommender systems. Despite notable progress in modeling user preferences from historical behaviors, two key challenges persist. First, exsiting discriminative paradigms focus on matching candidates to user history, often overfitting to historically dominant features and failing to adapt to rapid interest shifts. Second, a critical information chasm emerges from the point-wise ranking paradigm. By scoring each candidate in isolation, CTR models discard the rich contextual signal implied by the recalled set as a whole, leading to a misalignment where long-term preferences often override the user's immediate, evolving intent. To address these issues, we propose GenCI, a generative user intent framework that leverages semantic interest cohorts to model dynamic user preferences for CTR prediction. The framework first employs a generative model, trained with a next-item prediction (NTP) objective, to proactively produce candidate interest cohorts. These cohorts serve as explicit, candidate-agnostic representations of a user's immediate intent. A hierarchical candidate-aware network then injects this rich contextual signal into the ranking stage, refining them with cross-attention to align with both user history and the target item. The entire model is trained end-to-end, creating a more aligned and effective CTR prediction pipeline. Extensive experiments on three widely used datasets demonstrate the effectiveness of our approach.

Figures (7)

• Figure 1: Illustration of two key challenges in CTR prediction. Top: Relevance-based models focus on long-term preference matching and fail to adapt to short-term intent shifts (e.g., from iPhone accessories to MacBook accessories, ). Bottom: Inconsistency between recall and ranking, where the recall stage retrieves intent-relevant items (e.g., photography-related), but the ranking stage favors items aligned with long-term interests (e.g., history books).
• Figure 2: Overview of the proposed GenCI framework. The left part illustrates the generative module that constructs user interest cohorts, where "HISC" generates these cohorts via hierarchical quantization. The middle part details the Hierarchical Candidate-Aware Intent Modeling, which refines these cohorts. The right part showcases multi-intent fusion and jointly optimized with self-supervision for CTR prediction.
• Figure 3: Performance comparison w.r.t. different ensemble approaches applied in the HCAIM module.
• Figure 4: Visualization of different interest distribution. Short-term intents cluster tightly around target items, capturing immediate context, while long-term intents are broadly distributed, representing generalized preferences.
• Figure 5: Performance of GenCI w.r.t. number of Layers