Emotion-Based Call Summarization With Relevancy Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning summarization techniques require significant processing resources and large labeled training datasets, making them impractical for computers with limited computing resources and difficult to capture all important aspects of text for summaries.
Innovation Solution
A machine learning pipeline that integrates sentiment analysis to filter and rank text utterances, reducing the size of text corpuses and enabling accurate identification of relevant utterances for extractive summaries, thereby generating more nuanced and comprehensive summaries at a fraction of the traditional computational cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning summarization techniques are used, then summarization accuracy is improved, but processing resources and computational cost increase significantly
Solution Approach 1:
The patent segments the text processing task into two distinct stages: (1) sentiment analysis stage that processes the entire text to identify emotional patterns and relevant segments, and (2) summarization stage that processes only the identified relevant segments. This segmentation reduces the computational burden on the summarization model by limiting it to a subset of the original text, thereby maintaining accuracy while reducing processing resources.
Solution Approach 2:
The patent applies preliminary sentiment analysis before the summarization process to pre-process the text and identify relevant utterances. This preliminary action filters out irrelevant information before it reaches the summarization model, reducing the amount of data the model needs to process and thereby reducing computational cost while maintaining summarization quality.
2Manufacturing precision
If deep learning approaches are used for summarization, then summarization quality is improved, but processing complexity increases exponentially with text size
Solution Approach 1:
The patent divides the text processing into two stages where the first stage (sentiment analysis) handles the entire text and identifies relevant segments, while the second stage (summarization) processes only those segments. This segmentation reduces the input size to the summarization model exponentially, thereby reducing processing complexity while maintaining quality.
Solution Approach 2:
The patent extracts only the relevant text segments that contain emotionally significant information and uses these extracted segments as input to the summarization model. By taking out only the necessary portions of the text rather than processing the entire text, the system reduces processing complexity while maintaining summarization quality.
3Quantity of substance
If traditional summarization models are used, then comprehensive summaries are generated, but large labeled training datasets are required
Solution Approach 1:
The patent introduces sentiment analysis as an intermediary component that bridges the gap between raw text and summarization. This intermediary processes the text to identify relevant segments based on emotional content, thereby reducing the amount of training data needed for the summarization model to achieve comprehensive summaries. The intermediary handles the complex task of identifying relevant information, allowing the summarization model to focus on generation rather than content selection.
Data Source
AI summary
Embodiments of the present disclosure provide systems and methods for emotion-based call summarization. One method may include receiving an emotion prediction vector for an utterance text segment from a transcript data object, the emotion prediction vector comprising a plurality of emotion prediction scores respectively corresponding to a plurality of emotion identifiers; generating a domain-specific relevancy prediction for the utterance text segment based on a category-relevant subset of the plurality of emotion prediction scores that correspond to one or more category-specific emotion identifiers of the plurality of emotion identifiers associated with a domain-specific summarization category; identifying the utterance text segment as a relevant utterance from the transcript data object based on a comparison between the domain-specific relevancy prediction and a relevancy threshold; and initiating a performance of a machine learning summarization operation based on the utterance text segment.