Attention-Based Summarization Interpretation With Source-Text Cues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users struggle to understand how summarization models decide which portions of the source text to summarize and which to disregard, leading to a lack of confidence and trust in the model outputs.
Innovation Solution
A computerized method that analyzes attention values from summarization models to identify primary tokens in the source text, calculates semantic similarity with summary tokens, and generates visual cues to highlight relevant portions, enhancing user understanding and trust.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If summarization models automatically generate summaries without human intervention, then productivity is improved, but interpretability deteriorates
Solution Approach 1:
The patent introduces attention values as an intermediary mechanism that mediates between the automated summarization process and user interpretation. These attention values serve as a bridge, revealing which input tokens influenced which output tokens without requiring manual review of the source text, thus maintaining automation while improving interpretability
Solution Approach 2:
The system implements feedback by providing visual cues that highlight input tokens corresponding to selected summary portions. This feedback loop allows users to verify model decisions by seeing which source text segments were most influential, increasing trust while preserving automated operation
2Reliability
If visual cues highlight relevant portions of source text, then user confidence is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the most relevant information for interpretation by selecting a limited set of primary tokens based on attention values. Instead of displaying all possible connections, it extracts and highlights only the top contributing tokens, building user confidence without requiring complex visualization of the entire attention matrix
Solution Approach 2:
The system applies local quality by providing detailed interpretation information only at specific locations where users need it most - namely, by highlighting specific tokens in the source text that correspond to selected summary portions. This localized approach builds confidence without overwhelming the user with system-wide complexity
Data Source
AI summary
The disclosure herein describes interpreting attention-based decisions of summarization outputs generated by a deep learning model. A decision interpretation model obtains attention values defining connections between input tokens associated with a source text and output tokens for a selected portion of a summary associated with the source text. The input tokens having the highest attention values indicating the strongest connections between the input tokens of the source text and an output token of the summary are selected as primary tokens. A semantic similarity between the primary tokens for each attention head and an output token is calculated. The model selects the primary tokens having the closest semantic similarity with the summary portion. A visual cue is generated on or within a portion of the source text corresponding to the primary tokens. The visual cue identifies dominant words in the source text used to explain the summary portion.