Act2Intention: Benchmark for Active Mobile Agents via User Intention Inference
A new framework and benchmark called Act2Intention has been launched by researchers to create active mobile agents capable of deducing user intentions from GUI interactions. This initiative fills a void in existing mobile GUI agents, which tend to be reactive and lack a thorough understanding-prediction-execution cycle. Act2Intention combines the processes of comprehending and anticipating user intentions with decision-making execution. The accompanying benchmark, Act2Intention Bench, features 72,511 intentions and more than 700,000 actions from 52 applications, marking the first standard for assessing proactive agents through ongoing intention-action pathways. The Act2Intention Agent aims to deliver proactive services through Proactive-oriented Intention Understanding and Personalized Proactive Intention Prediction. The research paper can be found on arXiv under ID 2608.14132.
Key facts
- Act2Intention is a framework for active mobile agents.
- It integrates understanding, predicting user intentions, and executing decisions.
- Act2Intention Bench includes 72,511 intentions and over 700,000 actions.
- The benchmark spans 52 apps.
- It is the first benchmark for proactive agents via intention-action trajectories.
- The Act2Intention Agent provides proactive services.
- The paper is on arXiv with ID 2608.14132.
- The research focuses on multimodal large language models (MLLMs).
Entities
Institutions
- arXiv