Explainable Text Classification With Context And Reason Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional text classifiers in social media platforms rely on unsupervised learning and struggle to provide clear reasons for classifying textual content, leading to a need for improved understanding and trust in automated moderation decisions.
Innovation Solution
A text classifier system that includes a multi-head attention module, a knowledge base, and a reason classifier to determine the context and reasons behind the classification, providing explanations for the classification results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional text classifiers use unsupervised learning to classify text, then classification speed and automation are improved, but the ability to provide clear reasons for classification decisions deteriorates
Solution Approach 1:
The system segments the classification process into two distinct parts: (1) an unsupervised text classifier that performs automated classification, and (2) a separate explanation generator that produces human-interpretable reasons for the classification. This segmentation allows each component to specialize - the classifier focuses on accuracy and speed while the explanation generator focuses on interpretability, resolving the contradiction between automation and interpretability
Solution Approach 2:
The patent introduces an intermediary explanation generation system that acts as a bridge between the automated classifier and the user. This intermediary takes the classifier's output and generates natural language explanations, thereby preserving the automation benefits while adding the missing interpretability layer
2Reliability
If human moderators review comments to ensure accurate classification, then classification accuracy and trustworthiness are improved, but processing time and operational complexity increase
Solution Approach 1:
The system applies partial human intervention by using automated classifiers for the majority of cases and reserving human moderator review only for borderline or uncertain cases. The explanation generator provides sufficient information for most decisions, allowing human moderators to focus only on challenging cases, thus maintaining high reliability while reducing overall processing time
Solution Approach 2:
The system implements a feedback mechanism where explanation quality and classification confidence are continuously evaluated. When explanations are insufficient or confidence is low, the system automatically routes cases to human moderators, creating a feedback loop that improves overall system reliability while minimizing human involvement time
3Loss of information
If more context is extracted to improve explanation quality, then user understanding and trust are improved, but computational complexity and processing time increase
Solution Approach 1:
The explanation generation system applies local quality by focusing computational resources on extracting and explaining only the most relevant context segments rather than analyzing the entire text uniformly. It identifies key phrases and sentences that contribute most to the classification decision and generates explanations focused on these local regions, improving user understanding while controlling computational complexity
Data Source
AI summary
Apparatuses and methods are provided for classifying textual content using a text classifier for determining to which class the textual content belongs. After classification, the text classifier provides the classification result and a context relevant to the classification result to an explanation system. The explanation system predicts, from the classification result and the context relevant to the classification result, one or more reasons behind the classification result.