Attending to the Core: Core-Task Attention for Recommendation

Abstract

Industrial recommender systems typically support multiple business objectives through the integration of many specialized models. However, each individual model is usually optimized for a single primary target, such as conversions or purchases. For example, in advertising systems, models are often trained for OCPX-style objectives that tightly couple target prediction with bidding and revenue optimization. Since these target signals are often extremely sparse, multi-task learning (MTL) is widely adopted to leverage denser auxiliary tasks for additional supervision. However, existing MTL approaches typically pursue balanced joint optimization across tasks, which may introduce task interference and degrade the performance of the core task. To address this limitation, we propose Core Task Attention (CTAtt), a core-task-centric method that employs a novel attention mechanism to adaptively aggregate informative representations from auxiliary tasks into the core task’s prediction tower. CTAtt consists of two complementary attention pathways: (1) intra-sample attention, which models instance-level interactions among auxiliary tasks to produce context-aware fusion signals, and (2) inter-sample attention, which assesses each auxiliary signal’s global reliability by comparing its prediction score against the population distribution. These two pathways are fused through a lightweight gating mechanism to enrich the representation for the core task. Notably, CTAtt achieves strong performance even when built upon a simple Shared-Bottom architecture (CTAtt-SB). We evaluate CTAtt-SB on three public recommendation datasets, where it consistently outperforms strong MTL baselines. Moreover, online A/B tests on a leading short-video platform show that CTAtt significantly enhanced platform revenue and advertiser value by 3.3% and 2.3%, respectively. Our code is available here1.