Agent-based task processing method and apparatus, and electronic device
By training a large language model using positive and negative historical trajectory data, the adaptability and robustness of graphical user interface agents in the face of disturbances and non-standard tasks are solved, enabling the generation of correct operations in new environments and improving the processing capabilities of the agents.
CN122152477APending Publication Date: 2026-06-05VIVO MOBILE COMM CO LTD
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202610352838.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-05
AI Technical Summary
Technical Problem
Existing graphical user interface agents are prone to getting stuck in repetitive errors or operational loops when faced with disturbances, delays, or non-standard tasks, exhibiting low adaptability and robustness.
Method used
By training a large language model using positive and negative historical trajectory data, correct operational information can be generated and erroneous operations can be avoided, thereby improving the adaptability and robustness of the agent.
Benefits of technology
When faced with new environments or tasks, it can generate correct operational information, improving the adaptability and robustness of the agent in handling tasks.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122152477A_ABST
Abstract
The application discloses an agent-based task processing method and device and electronic equipment, and belongs to the technical field of artificial intelligence. The method comprises the following steps: receiving first task information, wherein the first task information is used for instructing an agent to execute a first task; searching in a positive trajectory library and a negative trajectory library respectively to obtain positive historical trajectory data and negative historical trajectory data matched with the first task information; the positive historical trajectory data in the positive trajectory library and the negative historical trajectory data in the negative trajectory library are obtained by reasoning task information samples through a large language model; generating prompt words according to the first task information, the positive historical trajectory data matched with the first task information and the negative historical trajectory data matched with the first task information; inputting the prompt words into the large language model to obtain first operation information output by the large language model, wherein the first operation information comprises N operations for executing the first task, and N is a positive integer; and executing the first task according to the first operation information through the agent.
Need to check novelty before this filing date? Find Prior Art