Intent Authoring Using Weak Supervision and Co-Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Customer contact center systems face challenges in accurately interpreting human intent from varied customer queries, as existing methods require labor-intensive annotation of utterances and often result in sub-optimal intent recognition performance due to limited labeled examples.
Innovation Solution
A method using weak supervision and co-training with propagation operations and learning algorithms is applied to a selected set of labeled conversational logs, allowing for efficient training of automated response systems by generalizing labels to cover linguistic variability and expanding the scope of intent recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If labor-intensive annotation of utterances is used to train automated response systems, then intent recognition performance can be improved, but the time and resources required for data preparation increase significantly
Solution Approach 1:
The system performs preliminary clustering of conversational logs into intent groups before annotation, organizing the data structure in advance. This preliminary action reduces the complexity of subsequent annotation tasks and enables more efficient training data preparation without compromising intent recognition accuracy
Solution Approach 2:
The system uses semi-supervised learning where the automated response system trains itself by leveraging both labeled and unlabeled conversational logs. The propagation operations automatically extend labels from labeled examples to unlabeled examples, reducing dependency on manual annotation while maintaining training effectiveness
2Productivity
If a small subset of labeled conversational logs is used for training, then the training process becomes faster and more efficient, but the intent recognition coverage and accuracy may be insufficient
Solution Approach 1:
The system introduces propagation operations as an intermediary mechanism that bridges the gap between labeled and unlabeled data. These operations propagate intent labels from the small labeled subset to the larger unlabeled corpus, effectively expanding the training coverage while maintaining training efficiency
Solution Approach 2:
The system employs co-training with multiple classifiers that learn from different feature representations of the conversational logs. By changing the parameters and perspectives through which the data is analyzed, the system achieves more robust intent recognition coverage even with limited labeled examples
3Measurement precision
If propagation operations and learning algorithms are applied to expand labeled data, then intent accuracy increases, but the complexity of the training process increases
Solution Approach 1:
The training process is segmented into distinct phases: initial labeling of a small subset, propagation of labels to expand the labeled corpus, and iterative co-training of multiple classifiers. This segmentation makes the complex process more manageable and allows for better control over each stage's contribution to the final intent accuracy
Data Source
AI summary
A combination of propagation operations and learning algorithms is applied, using a selected set of labeled conversational logs retrieved from a subset of a plurality of conversational logs, to a remaining corpus of the plurality of conversational logs to train an automated response system according to an intent associated with each of the conversational logs. The combination of propagation operations and learning algorithms may include defining the labels by a user for the selected set of the subset of the plurality of conversational logs; training a probabilistic classifier using the defined labels of features of the selected set, wherein the probabilistic classifier produces labeling decisions for the subset of conversational logs; weighting the features of the selected set in a model optimization process; and/or training an additional classifier using the weighted features of the selected set and applying the additional classifier to the remaining corpus.