Small language models improve user behavior prediction on video platforms
OmniSimulator: Aligning Small Language Models for Authentic Heterogeneous Behavior Modeling
Computation and LanguageArtificial IntelligenceMachine Learning
Summary
Simulating how real users behave online is hard, especially with smaller AI models that often just guess patterns. The authors propose OmniSimulator, a new way to teach smaller language models to better predict real user choices by showing them detailed, structured reasoning from real user data. This method helps the models think more about why users act certain ways rather than just copying surface-level patterns. Tests show that these smaller models get much better at mimicking user behavior, reaching performance close to larger, more complex models but easier to run on modest computers.
What this means in practice
- •For interactive system developers: Improve user simulators in video apps to better predict user engagement and inform system design decisions.
- •For chatbot and assistant builders: Create more realistic user simulations for testing and refining conversational AI in resource-constrained environments.
Authors
Jiawei Chen, Ruoxi Xu, Boxi Cao, Zhuoqun Li, Ruotong Pan, Yunfei Zhang, Zhirui Yang, Jingyan Chen, Tingting Gao, Han Li, Yaojie Lu, Xianpei Han, Le Sun, Xiangyu Wu, Hongyu Lin
Abstract
Large language models (LLMs) have shown great potential as general-purpose user simulators for interactive systems. Prior studies have demonstrated that training on high-quality data can substantially improve model reasoning and downstream task performance. However, the upper bound of user simulation is still constrained by the model's reasoning paradigm: native chain-of-thought often degenerates into shallow pattern matching and fails to adequately model the behavioral factors underlying real user decisions. In this paper, we propose OmniSimulator, a novel training framework designed to improve the predictive ability of small LLMs for realistic user simulation on short-video platform environments. Specifically, OmniSimulator organizes behavior-oriented reasoning through a structured schema with five dimensions, carefully selects data from real user-behavior scenarios, and introduces a teacher model to generate high-quality reasoning traces under the proposed schema, helping small models learn behavior-relevant reasoning patterns rather than relying on superficial behavioral correlations. Based on this process, we construct 9,881 high-quality training instances from three months of historical interaction trajectories of 200 real users, averaging about 50 distilled examples per user. Experimental results show that OmniSimulator improves the average score of Qwen3-8B by 34.33%, achieves performance competitive with Claude-Sonnet-4.5, and supports smaller-scale, self-hosted deployment. These results suggest that learning structured behavior-oriented reasoning patterns is a promising direction towards scalable and high-fidelity LLM-based user simulation.