Self-Learning Intent Recognition System for Speech Dialogue
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Intention recognition technologies in intelligent speech dialogue systems are overly dependent on the quality of rules or model training data and lack self-learning capabilities, requiring continuous manual intervention to maintain accuracy.
Innovation Solution
An intention recognition method with self-learning capability that acquires user expressions, performs preliminary recognition, and dynamically adjusts strategies based on historical data features, allowing for real-time updates and improved accuracy without relying on high-quality rule configurations or training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional intention recognition methods using rules or model training data are used, then the system can perform intention recognition, but the accuracy is overly dependent on the quality of rules or training data and requires continuous manual intervention
Solution Approach 1:
The system automatically queries historical data, performs feature extraction, and updates recognition strategies without manual intervention. The self-learning system autonomously processes prediction data, extracts features from historical intention recognition data, and dynamically adjusts recognition strategies based on extracted features, eliminating the need for continuous manual rule configuration or training data updates
Solution Approach 2:
The system establishes a feedback loop where prediction data from intention recognition is continuously fed back to the self-learning system. The self-learning system extracts features from this feedback data, analyzes patterns, and dynamically adjusts recognition strategies, creating a closed-loop system that continuously improves accuracy through automatic feedback processing
2Reliability
If manual intervention is used to maintain and provide accuracy of intention recognition, then the accuracy can be maintained, but the system lacks self-learning capability and requires continuous human involvement
Solution Approach 1:
The self-learning system autonomously performs all tasks previously requiring manual intervention. It automatically queries historical data based on prediction results, extracts relevant features, analyzes patterns, and updates recognition strategies without human involvement, completely eliminating the time loss associated with manual maintenance
Solution Approach 2:
The system pre-queries historical data and pre-extracts features before they are needed for recognition. The self-learning system continuously prepares recognition strategies by analyzing historical patterns in advance, so when new prediction data arrives, the system can immediately apply optimized strategies without waiting for manual preparation
Data Source
AI summary
An intent recognition method having a self-learning capability includes the following steps: acquiring a user expression, and recognizing a voice as a corresponding text; performing preliminary intent recognition on the user expression, and outputting candidate intents; acquiring historical data feature parameters of the candidate intents; on the basis of a pre-set rule strategy, deciding whether to directly output a final recognized intent, and on the basis of the feature parameters of each intent, performing rule computation, and outputting a final recognized intent; submitting prediction data of the final recognized intent and the candidate intents from the intent recognition process to a self-learning system, and performing self learning and indicator parameter data updating. The present disclosure is able to perform self learning on the basis of the feature distribution in historical data of intent recognition and dynamically adjust intent recognition strategies.