Long Dialog Summarization Using Recursive Topic-by-Topic Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abstractive summarization techniques fail to generate coherent and context-grounded summaries for long, unstructured dialogs, resulting in disjointed and irrelevant summaries.
Innovation Solution
A machine learning recursive abstractive model is trained to generate summaries of long form dialogs by considering each preceding sentence when generating the next, using a divide and conquer approach to improve coherence and relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional summarization techniques are used to summarize long dialogs by individually summarizing portions, then the summarization process is simpler and faster, but the summaries become disjointed and lose overall context coherence
Solution Approach 1:
The patent divides the long dialog into multiple segments or portions, processes each segment individually through the summarization model, and then combines the segment summaries into a final summary. This segmentation approach allows the model to handle long dialogs effectively while maintaining context coherence through the recursive processing of segments.
Solution Approach 2:
The patent implements a recursive summarization architecture where segment summaries are nested within the overall dialog context. Each segment summary is generated considering the context of previous segments, creating a nested structure where local segment information is integrated into the global dialog summary, maintaining both detail and overall coherence.
2Measurement precision
If individual portions of long dialogs are summarized separately, then processing time and computational resources are reduced, but the summaries become inaccurate and emphasize irrelevant portions
Solution Approach 1:
The patent performs preliminary processing by dividing the dialog into segments and generating segment summaries in advance. These segment summaries are then used as input for the final summarization step, allowing the model to focus computational resources on integrating context rather than processing raw text from scratch, thus improving accuracy while managing processing time.
Solution Approach 2:
The patent maintains continuous context flow by feeding previous segment summaries into the processing of subsequent segments. This continuous action ensures that the summarization process maintains awareness of the overall dialog context throughout processing, improving accuracy without requiring reprocessing of entire dialog at each step.
Data Source
AI summary
Various embodiments of the present disclosure provide summarization techniques for summarizing complex documents, such as long unstructured call transcripts. The summarization techniques include generating a plurality of interaction topics for an interaction transcript and iteratively summarizing each interaction topic based on a preceding partial summary for the interaction transcript that corresponds to a preceding interaction topic that precedes the interaction topic in the interaction transcript. An abstractive summary is generated using a recursive abstractive model that is trained using training data generated based on holistic similarity scores between interaction topics of a call transcript and summary sentences of a corresponding target summary.