AI Semantic Recognition Stabilized by Deterministic Negative Corpus Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human-computer dialog systems face instability due to fluctuations in model confidence during retraining, leading to inconsistent recognition of training corpora and poor developer experience.
Innovation Solution
An artificial intelligence-based semantic recognition method that stabilizes the model by extracting a negative corpus based on the encoding value of the training corpus using a mapping relationship, ensuring that the model remains unchanged when the training corpus is not modified.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the model is retrained multiple times, then the model can be optimized and improved, but the model confidence fluctuates significantly and recognition results become inconsistent
Solution Approach 1:
The patent applies preliminary action by extracting negative corpora in advance based on encoding values before model training begins. This pre-extraction ensures that the same negative corpora are used across multiple training iterations, preventing confidence fluctuations and maintaining consistent recognition results when the training corpus remains unchanged.
Solution Approach 2:
The patent uses encoding values as stable identifiers that copy the essential characteristics of training corpora without being affected by data ordering. This copying mechanism ensures that the mapping between training corpora and negative corpora remains consistent across different training sessions, resolving the instability issue.
2Ease of manufacture
If random negative corpora are extracted during each training session, then the training process is simple, but the model produces inconsistent results across different training sessions
Solution Approach 1:
The patent changes the parameter of negative corpus selection from random sampling to deterministic selection based on encoding values. This parameter change maintains the simplicity of the training process while ensuring that the same training corpus always maps to the same negative corpora, producing consistent model outputs across training sessions.
3Adaptability or versatility
If the training corpus ordering changes between training sessions, then data processing flexibility is improved, but the model recognition results become unstable
Solution Approach 1:
The patent introduces encoding values as an intermediary between the training corpus and negative corpus selection. This intermediary remains invariant to changes in data ordering, allowing flexible data processing while maintaining stable recognition results through consistent corpus mapping.
Data Source
AI summary
An artificial intelligence-based semantic recognition method, apparatus, and device. In the artificial intelligence-based semantic recognition method, a pre-trained semantic recognition model is trained by using a training corpus configured by a developer on a model training platform such as a Bot platform and a negative corpus provided on the model training platform, where the negative corpus is extracted by mapping an encoding value of the training corpus to a negative corpus set. Therefore, the negative corpus is extracted based on the encoding value of the training corpus, and a randomized method for generating the negative corpus is changed into a stable method.