Agent Task Orchestration for Complex User Intent Execution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing agents based on large language models are unable to process complex user intents due to limited recognition and manipulation abilities, leading to unsatisfactory user experiences.
Innovation Solution
A method and device that utilize a Large Language Model (LLM) to recognize user operation intents, split them into sub-tasks, and perform task orchestration to generate and execute a target text sequence, enhancing the ability to handle complex tasks by integrating cue words, logical reasoning, and API interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing agents based on large language models are used, then the system can process simple user intents, but the system cannot process complex user intents due to limited recognition and manipulation abilities
Solution Approach 1:
The patent segments complex user intents into multiple sub-tasks through hierarchical decomposition. The LLM analyzes the user's intent and breaks it down into smaller, manageable sub-tasks that can be independently processed and executed, thereby improving the system's ability to handle complex operations while maintaining reliability through structured task management
Solution Approach 2:
The patent introduces an intermediary component that bridges the LLM and the agent execution system. This intermediary handles task decomposition, sub-task orchestration, and coordination between different components, enabling the system to process complex intents by mediating between high-level user commands and low-level execution actions
2Adaptability or versatility
If the LLM model performs logical reasoning and task orchestration, then complex tasks can be processed, but the system complexity increases
Solution Approach 1:
The patent implements a universal task orchestration framework that handles multiple types of complex tasks through a single integrated system. The LLM model serves multiple functions including intent recognition, task decomposition, sub-task generation, and execution coordination, thereby reducing overall system complexity despite enhanced task processing capabilities
Solution Approach 2:
The patent employs a nested structure where complex tasks are decomposed into sub-tasks, which may further decompose into smaller sub-sub-tasks. This hierarchical nesting allows the system to manage complexity by organizing tasks at multiple levels, with each level handling appropriate granularity of operations
3Adaptability or versatility
If task orchestration is performed for multiple sub-tasks, then complex user operations can be executed, but the processing time increases
Solution Approach 1:
The patent performs preliminary analysis of user intents to identify and decompose tasks before execution. By pre-processing the intent recognition and task decomposition phases, the system prepares execution plans in advance, allowing for more efficient runtime execution of complex operations and reducing overall processing time
Data Source
Figure 1~3
Figure 4~5
AI summary
A method for business processing based on an agent includes: recognizing (202) an operation intent of a user for a host target of the agent and generating a target text indicative of the operation intent of the user for the host target based on a recognition result; generating (204) a cue word based on the target text, inputting the cue word into an LLM model carried on the agent, performing, by the LLM model, logical reasoning based on the cue word, splitting the operation intent into at least one sub-task, and performing a task orchestration for the at least one sub-task to generate a target text sequence including a task description text corresponding to each sub-task; obtaining (206) the target text sequence output by the LLM model, and performing each sub-task contained in the target text sequence, to perform an operation corresponding to the operation intent.