Call Class Prediction Using MaxEnt and CRF Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Call center agents face inefficiencies due to lack of prior information about incoming calls, leading to slower processing and misunderstandings, which can be addressed by predicting the class of future customer calls to enhance interactions and staffing strategies.
Innovation Solution
A system using a graphical user interface that analyzes prior call information through tokenization and employs a maximum entropy model and conditional random field classifier to predict call classes, accounting for agent workflows and noisy data, with adaptive retraining based on accuracy thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If agents handle calls without prior information about the call subject, then they can maintain operational simplicity, but call handling time increases and productivity decreases
Solution Approach 1:
The system performs preliminary analysis of customer data before the call occurs, predicting the call subject and class using machine learning models. This advance preparation provides agents with relevant information before they answer the call, eliminating the need to interpret problem formulations during the call and significantly reducing call handling time.
2Measurement precision
If the system uses complex machine learning models to predict call classes, then prediction accuracy improves, but system complexity and computational requirements increase
Solution Approach 1:
The prediction system is divided into multiple independent components: data preprocessing module, feature extraction module, machine learning model module, and integration module. Each component handles a specific aspect of the prediction task, allowing for optimized development and deployment of individual modules while maintaining overall system accuracy.
Solution Approach 2:
The patent introduces an intermediary layer that bridges complex machine learning models and the simple call center interface. This intermediary handles model training, prediction generation, and result formatting, shielding the complexity of advanced algorithms from the end users while delivering accurate predictions through a user-friendly interface.
3Measurement precision
If the system leverages extensive customer data for predictions, then prediction quality improves, but data privacy and security concerns increase
Solution Approach 1:
The system extracts only the essential features and patterns needed for prediction from extensive customer data, rather than storing or processing all raw data. This extraction approach maintains prediction quality by preserving key information while minimizing data storage requirements and reducing privacy risks associated with retaining sensitive customer information.
Data Source
AI summary
A system and method for predicting the class of future customer calls to a call center. Saved call data is analyzed using a robust tokenizer of a computerized device. The tokenizer transforms a sequence of characters in a call summary field of the saved call data into a sequence of tokens. Tokenized call data is produced. Multiple maximum entropy (MaxEnt) models are created based on the tokenized call data, using the computerized device. The MaxEnt models produce a probability distribution of all classes for a next call to a call center. A conditional random field (CRF) classifier is trained with the MaxEnt models and information from the saved call data, using the computerized device. The CRF classifier uses chronologically ordered sequences of prior calls to the call center and predicts a class for a new call to the call center based on the saved call data. A call class prediction is produced for the new call received from a returning customer based on the CRF classifier and the MaxEnt model.