Transcript Topic Assignment Using Utterance Segmentation Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio-to-text systems fail to distinguish between key phrases uttered by a user and those uttered by an entity, leading to less valuable transcripts, wasted processing resources, and unnecessary storage due to inability to differentiate importance levels.
Innovation Solution
A system that extracts key phrases from separate portions of a transcript, assigns them to topics, and calculates topic frequencies, generating a final set of topics based on predefined frequency thresholds to prioritize important utterances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If key phrases are extracted from all portions of the transcript without differentiation, then completeness of topic coverage is improved, but processing resources and storage requirements increase unnecessarily
Solution Approach 1:
The patent segments the transcript into two distinct portions: user utterances and entity utterances. By applying different key phrase extraction processes to each segment, the system prioritizes user utterances which contain more valuable information, while still processing entity utterances to ensure topic coverage completeness. This segmentation resolves the contradiction by maintaining information completeness while optimizing resource allocation.
2Loss of information
If key phrases from both user and entity utterances are processed equally, then comprehensive topic identification is improved, but storage requirements increase due to redundant data
Solution Approach 1:
The patent applies local quality by treating user utterances and entity utterances differently in the key phrase extraction process. User utterances receive prioritized processing with higher importance weight, while entity utterances are processed with lower priority. This differential treatment ensures that storage resources are allocated efficiently based on the actual value and importance of the information source.
3Adaptability or versatility
If all key phrases are assigned to topics without frequency filtering, then topic diversity is improved, but the value of the transcript is reduced due to inclusion of low-importance phrases
Solution Approach 1:
The patent changes the parameter of topic inclusion by introducing a frequency threshold criterion. Topics are only included in the final set if they meet a minimum frequency threshold, which filters out low-importance topics that would dilute transcript value. This parameter change resolves the contradiction by maintaining meaningful topic diversity while excluding noise, thereby preserving transcript value.
Data Source
AI summary
In some implementations, a system may obtain a transcript that includes interactions between a user and an entity. The system may extract a first quantity of key phrases from a first portion of the transcript that corresponds to an entirety of the transcript and may extract a second quantity of key phrases from a second portion of the transcript that corresponds to a subset of the entirety of the transcript. The system may assign one or more key phrases to one or more topics, and may calculate a topic frequency that indicates a total quantity of key phrases associated with the topic. The system may generate a third set of topics that includes one or more topics having a topic frequency that satisfies a topic frequency threshold.