LLM Intent Recognition with Candidate Filtering for Complex Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing conversational human-computer interaction systems face challenges in accurately recognizing user intents due to the complexity of natural language processing, particularly in scenarios involving multiple intents or interactions.
Innovation Solution
A method and apparatus utilizing a large language model (LLM) for intent recognition, involving the steps of obtaining query statements, matching them with preset intents and descriptive information, generating prompt information, and determining target intents through the LLM, with optional fine-tuning using sample datasets and historical interaction data to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional intent recognition methods are used, then the system is simpler to implement, but the accuracy of intent recognition deteriorates due to the complexity of natural language processing
Solution Approach 1:
The patent segments the intent recognition process into multiple distinct modules: a candidate intent generation module that identifies potential intents, a prompt information generation module that creates structured input, and an LLM-based determination module that makes the final classification. This segmentation allows each module to specialize in a specific task, improving overall accuracy while making the complex system more manageable and maintainable through modular architecture.
Solution Approach 2:
The patent introduces prompt information as an intermediary element between the candidate intent generation module and the LLM determination module. This prompt information serves as a structured bridge that transforms unstructured natural language queries into a format that the LLM can effectively process, thereby enhancing recognition accuracy without requiring direct complex interaction between the natural language input and the LLM.
2Adaptability or versatility
If comprehensive intent matching is performed, then the coverage of recognized intents improves, but the time consumption increases due to processing multiple intents and historical context
Solution Approach 1:
The patent performs preliminary action by generating candidate intents and their corresponding prompt information before the LLM makes the final determination. This pre-processing step filters and structures the data in advance, allowing the LLM to focus only on evaluating pre-identified candidates rather than processing all possible intents from scratch, thereby reducing overall processing time while maintaining comprehensive intent coverage.
Solution Approach 2:
The patent generates multiple candidate intents (excessive action) beyond what might be strictly necessary, then uses the LLM to select the most appropriate one. This approach ensures that no potential intent is missed (improving coverage) while the LLM's efficient selection process prevents linear increase in processing time, as the LLM can quickly evaluate and rank multiple candidates.
3Measurement precision
If LLM is used for intent recognition, then the accuracy of recognizing complex queries with multiple intents improves, but the computational resources required increase
Solution Approach 1:
The patent segments the computational workload by first using a lighter candidate generation module to identify potential intents, then using the LLM only for the final determination step. This segmentation allows the system to leverage the LLM's superior accuracy for complex multi-intent recognition while minimizing its usage to only the critical decision-making phase, thereby reducing overall computational resource consumption compared to using the LLM for the entire processing pipeline.
Data Source
AI summary
The present disclosure provides a method and an apparatus for intent recognition based on a large language model (LLM), an electronic device, and a storage medium, relating to a field of computer technology, specifically to a field of artificial intelligence technology, such as natural language processing and an LLM. A specific implementation solution is as follows: obtaining a query statement, a preset intent, and descriptive information of the preset intent; obtaining a first candidate intent corresponding to the query statement by matching the query statement with the preset intent and the descriptive information of the preset intent; generating first prompt information based on the query statement, the first candidate intent, and descriptive information of the first candidate intent; and determining a first target intent corresponding to the query statement from the first candidate intent by inputting the first prompt information into the LLM.