LLM Intent Recognition with Candidate Filtering for Query Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intent recognition systems face challenges in accurately identifying user intents in conversational human-computer interaction scenarios due to interference from irrelevant intents, leading to reduced accuracy and increased difficulty in intent recognition.
Innovation Solution
A method and apparatus utilizing a large language model (LLM) to obtain query statements, preset intents, and descriptive information, generate prompt information, and determine target intents by matching and fine-tuning the LLM based on candidate intents and historical interaction data, thereby reducing interference and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional intent recognition methods are used, then the system can process queries, but accuracy is reduced due to interference from irrelevant intents
Solution Approach 1:
The patent extracts and removes irrelevant intents from the candidate intent list before presenting to the LLM. By filtering out intents that do not match the query semantics (using cosine similarity thresholding and relevance scoring), the system eliminates harmful interference factors that would otherwise degrade recognition accuracy
Solution Approach 2:
The patent segments the intent recognition process into distinct stages: initial candidate generation, relevance filtering, and final recognition by LLM. This segmentation allows the system to handle different aspects of intent recognition separately, improving overall accuracy by addressing irrelevant intents in the filtering stage before they can interfere with the final recognition
2Measurement precision
If all candidate intents are presented to the LLM, then comprehensive coverage is achieved, but task difficulty increases and accuracy decreases
Solution Approach 1:
The patent extracts only the most relevant candidate intents (e.g., top 3-5) after filtering, removing the rest. This reduction in the number of candidate intents presented to the LLM simplifies the classification task while maintaining comprehensive coverage of actually relevant intents, thereby improving accuracy and reducing task difficulty
Solution Approach 2:
The patent applies different processing qualities to different candidate intents: highly relevant intents receive detailed analysis and are presented to the LLM, while less relevant intents are filtered out or given lower priority. This local quality approach ensures that the LLM focuses computational resources on the most important distinctions
Data Source
Figure 1~2
Figure 3~4
Figure 5
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.