Real-Time Transcript Topic Classification Using Attention-Weighted Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Processing large volumes of transcript data in real-time using traditional language models is impractical due to the computational resources required, especially for providing immediate feedback in communication sessions.
Innovation Solution
A method and system that utilize a smaller and faster classifier model to classify feedback from transcripts by processing utterances with attention values and encoder values to identify topic labels, allowing for real-time routing of communication sessions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional language models are used to process transcript data, then analysis accuracy is improved, but processing time and computational resources increase significantly
Solution Approach 1:
The patent segments the transcript processing task into multiple stages: initial filtering using a smaller model to identify relevant utterances, then applying the larger language model only to those selected segments. This segmentation allows the system to maintain high analysis accuracy on relevant portions while reducing overall processing time and computational resource usage.
Solution Approach 2:
The patent extracts and removes irrelevant portions of transcripts before applying the full language model analysis. By using a lighter initial model to filter out unnecessary content, the system extracts only the meaningful utterances that require detailed analysis, thereby reducing the time and computational resources needed while preserving accuracy on the extracted relevant content.
2Reliability
If larger language models with more parameters are used, then model capability and accuracy are improved, but the time required to generate output increases
Solution Approach 1:
The patent implements a dynamic model selection strategy where the system adjusts which model to use based on the specific task requirements and input characteristics. For routine filtering tasks, a smaller faster model is used, while for complex analysis tasks requiring high reliability, the larger model is deployed. This dynamic approach balances model capability with output generation speed.
Solution Approach 2:
The patent applies partial action by using the smaller model for the majority of the processing pipeline (filtering and preliminary analysis) and reserving the larger, more capable model only for critical tasks where its enhanced capability is truly necessary. This partial deployment of the larger model maintains high reliability where needed while preserving overall productivity.
3Speed
If real-time analysis of communication sessions is implemented, then responsiveness and user feedback time are improved, but computational resource requirements increase
Solution Approach 1:
The patent performs preliminary filtering and preprocessing of transcript data using computationally efficient methods before the main analysis. By pre-identifying relevant utterances and segments using a lighter model, the system prepares the data in advance, enabling real-time responsiveness while keeping computational resource consumption manageable during the actual analysis phase.
Data Source
AI summary
A method classifies feedback from transcripts. The method includes receiving an utterance from a transcript from a communication session and processing the utterance with a classifier model to identify a topic label for the utterance. The classifier model is trained to identify topic labels for training utterances. The topic labels correspond to topics of clusters of the training utterances. The training utterances are selected using attention values for the training utterances and clustered using encoder values for the utterances. The method further includes routing the communication session using the topic label for the utterance.