SmartAgent: Chain-of-User-Thought for Embodied Personalized Agent in Cyber World

The official repository of our paper "SmartAgent: Chain-of-User-Thought for Embodied Personalized Agent in Cyber World". We will release code and data upon paper notification.

Blogs 🤗

Find SmartAgent in Awesome-LLM-Reasoning with more brilliant works on LLM/MLLM reasoning!

Chain-of-User-Thought (COUT) Reasoning Paradigm

We formulate COUT to achieve embodied personalized agent training in terms of three stages of thought. In Thought #1, according to a user's instruction, an agent performs GUI actions to search for an item pool. In Thought #2 with seeing the pool, the agent reasons underlying requirements behind the original instruction, as implied by the previous actions. In Thought #3, based on the underlying thought, the agent recommends items within the pool to complete the user's instruction.

By leveraging user-oriented thoughts, this COUT could enable full-stage embodied personalized capabilities across various information systems.

SmartSpot Benchmark

The first environment supports full-stage embodied personalized evaluation.

SmartAgent Model

The capabilities of SmartAgent from basic embodied operations to personalized reasoning.

Two-stage training paradigm of SmartAgent.

Citation

Please consider citing our paper and staring this repo if you find SmartAgent helpful in your work, thanks!

@article{zhang2024smartagent,
      title={SmartAgent: Chain-of-User-Thought for Embodied Personalized Agent in Cyber World}, 
      author={Zhang, Jiaqi and Gao, Chen and Zhang, Liyuan and Li, Yong and Yin, Hongzhi},
      journal={arXiv preprint arXiv:2412.07472},
      year={2024}
}