Dialog Model Utterance Relevance Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing dialog systems face challenges in task completion efficiency due to training on dialog data that includes non-essential utterances, which are not necessary for task completion.
Innovation Solution
A method is introduced to identify and remove non-essential utterances from the training data by computing a relevance value for each utterance, using a threshold to determine its importance, thereby generating an adapted dialog model that focuses on essential information for task completion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If dialog systems are trained on complete dialog data including all utterances, then the model retains comprehensive information, but task completion efficiency deteriorates due to non-essential utterances
Solution Approach 1:
The patent extracts and removes non-essential utterances from the training dialog data by computing relevance values between utterances and task concepts. Only utterances with relevance values above a threshold are retained, thereby eliminating harmful information while preserving essential task-related content, resolving the contradiction between efficiency and information completeness
Solution Approach 2:
The patent applies different quality standards to different parts of the dialog data by selectively filtering utterances based on their individual relevance scores. Each utterance is evaluated locally against task concepts, and only those meeting the quality threshold are kept, creating a heterogeneous training set with varying levels of importance that optimizes both efficiency and completeness
2Reliability
If all utterances are retained in training data, then information completeness is maintained, but model complexity and processing time increase
Solution Approach 1:
The patent removes unnecessary components (non-essential utterances) from the training data through relevance-based filtering. This extraction process reduces the overall size and complexity of the dialog model while maintaining the essential information needed for reliable task completion, directly addressing the contradiction between reliability and complexity
3Productivity
If non-essential utterances are removed from training data, then task completion efficiency improves, but the risk of removing essential information increases
Solution Approach 1:
The patent implements a feedback mechanism where utterances are evaluated against task concepts using relevance computation. This feedback loop ensures that only utterances with sufficient relevance to task concepts are retained, providing a systematic way to verify that essential information is preserved while removing non-essential content, thus maintaining reliability while improving efficiency
Solution Approach 2:
The patent changes the parameter of utterance selection from binary inclusion to continuous relevance scoring. By computing relevance values and applying a threshold, the system dynamically adjusts which utterances are included based on their measured relevance to task concepts, optimizing the balance between efficiency improvement and reliability maintenance
Data Source
AI summary
A computer-implemented method identifies one or more concepts from a document, wherein the document comprises information on completing a task. The method computes a relevance value of an utterance with respect to completing the task using the one or more identified concepts. The method removes the utterance from a dialog model to be used for completing the task when the relevance value of the utterance is below a given threshold value.