Intention Identification Model Training via Label Reset
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In human-computer interaction systems, existing intention identification models face errors when answering questions without considering the context of the question, leading to incorrect intention recognition due to insufficient training data, where answers are misidentified across different questions.
Innovation Solution
The intention identification model training method involves weakening the training of target questions in initial rounds of model training to reduce the probability of incorrect intention identification, followed by resetting labels of identified intentions to improve accuracy, allowing the model to distinguish between intentions correctly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If the model is directly trained by using answer and intention without combining with question, then training data processing is simple, but the model identifies all answers as having the same intention resulting in identification error
Solution Approach 1:
The patent segments the training process into multiple rounds where different aspects are emphasized. In early rounds, the model learns basic answer-intention mappings. In later rounds, the model learns to differentiate intentions based on question-context combinations. This segmentation allows the system to progress from simple processing to accurate identification.
Solution Approach 2:
The patent applies preliminary action by first training the model with answer-intention pairs without question context to establish basic mappings. Then, in subsequent training rounds, question context is introduced to refine and differentiate intentions. This preliminary training phase prepares the model for more complex contextual learning.
2Measurement precision
If the model combines answer with question to determine intention, then intention identification accuracy improves, but the model incorrectly identifies answers as having same intention across different questions
Solution Approach 1:
The patent implements dynamics by making the training process adaptive across multiple rounds. The model dynamically adjusts its learning focus: early rounds emphasize answer-intention mapping while later rounds emphasize question-answer interaction. This dynamic training approach allows the model to learn both general patterns and specific contextual distinctions.
Solution Approach 2:
The patent uses feedback mechanisms where the model's identification results from early rounds inform subsequent training adjustments. By analyzing where the model makes errors in distinguishing intentions across different questions, the training process is refined to reduce these specific errors while maintaining overall accuracy.
3Productivity
If training emphasizes answer-intention mapping without question context, then training convergence is fast, but model generalization to new questions fails
Solution Approach 1:
The patent applies preliminary action by first establishing quick answer-intention mappings in early training rounds to achieve fast convergence. Then, in subsequent rounds, question context is gradually introduced to enhance generalization capability. This staged approach allows the model to converge quickly initially while still developing adaptability.
Solution Approach 2:
The patent implements periodic action through multi-round training where the emphasis shifts periodically between answer-intention mapping and question-context integration. This periodic variation in training focus allows the model to alternatingly improve convergence speed and generalization capability across different training phases.
Data Source
AI summary
Implementations of the present specification describe an intention identification model training method and apparatus, and an intention identification method and apparatus. According to the methods in the implementations, training of a target question can be weakened in the first several rounds of model training, and then an intention identification model obtained in the first several rounds of training can be used to identify intentions corresponding to answers that are to be distinguished. Further, the intention identification model is trained again after labels of these intentions are reset, so that the intention identification model obtained through training can also have a good identification effect on an answer to the target question, thereby improving accuracy of intention identification.