How Do Decoder-Only LLMs Perceive Users? Rethinking Attention Masking for User Representation Learning

TL;DR

This paper systematically studies how different attention masking strategies affect user representation learning in decoder-only LLMs, proposing a gradient-guided soft masking method to improve training stability and embedding quality.

Contribution

It introduces Gradient-Guided Soft Masking for better transition from causal to bidirectional attention in large-scale user modeling.

Findings

Gradient-Guided Soft Masking improves training stability.

Bidirectional attention yields higher-quality user embeddings.

Method outperforms baseline masking strategies on multiple benchmarks.

Abstract

Decoder-only large language models are increasingly used as behavioral encoders for user representation learning, yet the impact of attention masking on the quality of user embeddings remains underexplored. In this work, we conduct a systematic study of causal, hybrid, and bidirectional attention masks within a unified contrastive learning framework trained on large-scale real-world Alipay data that integrates long-horizon heterogeneous user behaviors. To improve training dynamics when transitioning from causal to bidirectional attention, we propose Gradient-Guided Soft Masking, a gradient-based pre-warmup applied before a linear scheduler that gradually opens future attention during optimization. Evaluated on 9 industrial user cognition benchmarks covering prediction, preference, and marketing sensitivity tasks, our approach consistently yields more stable training and higher-quality bidirectional representations compared with causal, hybrid, and scheduler-only baselines, while remaining compatible with decoder pretraining. Overall, our findings highlight the importance of masking design and training transition in adapting decoder-only LLMs for effective user representation learning. Our code is available at https://github.com/JhCircle/Deepfind-GGSM.