IRA: Adaptive Interest-aware Representation and Alignment for Personalized Multi-interest Retrieval

TL;DR

IRA tackles dynamic, multi-interest personalized retrieval by representing user interests as a set of Interest Units that accumulate updates and fade over time, and by aligning these units with documents through a specialized embedding model. It avoids reliance on click signals to mitigate temporal biases and enables near real-time adaptation without full retraining. The approach demonstrates strong, fine-grained personalization on real-world NAVER CAFE data, including an online A/B deployment that improved engagement metrics. These results argue for the practicality and scalability of IRA in large-scale industrial settings.

Abstract

Online community platforms require dynamic personalized retrieval and recommendation that can continuously adapt to evolving user interests and new documents. However, optimizing models to handle such changes in real-time remains a major challenge in large-scale industrial settings. To address this, we propose the Interest-aware Representation and Alignment (IRA) framework, an efficient and scalable approach that dynamically adapts to new interactions through a cumulative structure. IRA leverages two key mechanisms: (1) Interest Units that capture diverse user interests as contextual texts, while reinforcing or fading over time through cumulative updates, and (2) a retrieval process that measures the relevance between Interest Units and documents based solely on semantic relationships, eliminating dependence on click signals to mitigate temporal biases. By integrating cumulative Interest Unit updates with the retrieval process, IRA continuously adapts to evolving user preferences, ensuring robust and fine-grained personalization without being constrained by past training distributions. We validate the effectiveness of IRA through extensive experiments on real-world datasets, including its deployment in the Home Section of NAVER's CAFE, South Korea's leading community platform.

Figures (3)

• Figure 1: The overview of IRA pipeline. When the user clicks $d_1$ similar to the existing Unit $c_1$, $c_1$ is reinforced. When the user no longer clicks any documents similar to $c_3$, $c_3$ is gradually removed. The original was in Korean, but translated to English.
• Figure 2: Analysis of big Unit distribution (Left) and the number of Units (Right).
• Figure 3: Analysis of Unit pruning strategies (Left) and contextual text construction (Right).