Hierarchical Machine Learning for Call Transcription Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to efficiently analyze member calls for prescription drug plans due to the difficulty in determining specific reasons for calls using broad categories, leading to a time-consuming and resource-intensive manual detailed root cause analysis.
Innovation Solution
A computer system that integrates transcription and agent call log data, uses natural language processing and machine learning models like LDA to classify and sub-classify call data entries, enhancing the analysis of member calls.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual detailed root cause analysis is performed for each call, then analysis accuracy is improved, but time consumption and resource intensity increase
Solution Approach 1:
The system performs preliminary automated classification of calls using machine learning models before detailed analysis. Call transcriptions are pre-processed and categorized into broad topics, which prepares the data for more efficient subsequent analysis and reduces the manual effort required for each individual call
Solution Approach 2:
The analysis process is segmented into multiple hierarchical levels: broad topic classification, sub-topic classification, and detailed root cause analysis. This segmentation allows automated processing at higher levels and manual intervention only at specific detailed levels, reducing overall time consumption while maintaining accuracy
2Productivity
If broad categories are used for call analysis, then processing efficiency is improved, but ability to detect trends and anomalies deteriorates
Solution Approach 1:
The system adds dimensional depth to call analysis by implementing a hierarchical classification structure with multiple levels (broad topics, sub-topics, and detailed categories). This multi-dimensional approach allows efficient broad categorization while preserving detailed information necessary for trend detection through the combination of automated and manual analysis at different hierarchical levels
3Speed
If automated classification systems are implemented, then processing speed is improved, but classification precision deteriorates
Solution Approach 1:
The system introduces hybrid automated-manual classification as an intermediary approach. Automated machine learning models perform initial classification at high speed, and human analysts review and refine classifications for complex or ambiguous cases. This intermediary manual review step maintains high processing speed while improving classification precision where needed
Data Source
AI summary
A computer system includes memory hardware configured to store a transcription database, a call database, and instructions. The transcription database includes multiple call transcription data entries. The call database includes multiple agent call log data entries. The computer system executes instructions including joining the call transcription data entries with the agent call log data entries according to associated timestamps. The instructions include, for at least one of the set of joined call data entries, generating an input vector for an unsupervised machine learning model. The instructions include supplying the input vector to the model to assign an output topic classification of the model to the joined call data entry associated with the input vector, and supplying the input vector to at least one sub-topic model associated with the output topic classification to assign one or more sub-topic output classifications to the joined call data entry associated with the input vector.