Call Classification via Rule and Probability Arbitration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing call center systems face challenges in accurately categorizing calls due to the limitations of pre-defined binary rules, which require extensive tuning to avoid false positives or negatives, and are not effective for call reasons without clear associated actions, necessitating a more efficient method for categorization.
Innovation Solution
A computer system comprising a rules module, a probabilistic module, and an arbitration module that determine the probability of a call belonging to a category using both rules and probabilistic models, allowing for soft predictions and continuous learning from correctly classified calls to improve categorization accuracy over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If pre-defined binary rules are used for call categorization, then the categorization process is simple to implement, but the accuracy is insufficient and extensive tuning is required to avoid false positives or negatives
Solution Approach 1:
The system segments the categorization process into multiple independent modules: a rules module that applies predefined binary rules to generate initial predictions, and a probabilistic module that uses machine learning models to analyze patterns in the data. Each module operates independently with its own logic, allowing the system to combine simple rule-based approaches with complex probabilistic analysis to achieve both ease of implementation and high accuracy
Solution Approach 2:
The system creates a composite categorization approach by combining two different methodologies: deterministic binary rules and probabilistic machine learning models. The rules module provides interpretable, easy-to-understand categorization for clear-cut cases, while the probabilistic module handles ambiguous cases and learns from historical data. This composite approach leverages the strengths of both methods to achieve high accuracy without requiring extensive tuning of a single complex system
2Device complexity
If pre-defined binary rules are used for call categorization, then the system structure is simple, but extensive time is required for tuning to avoid false positives or negatives
Solution Approach 1:
The probabilistic module is trained in advance on historical call data to learn patterns and relationships between call features and categorizations. This preliminary training action allows the system to capture complex patterns that would require extensive manual tuning of binary rules. Once trained, the probabilistic model can be applied directly to new calls without requiring additional tuning, significantly reducing the time investment needed for deployment and maintenance
Solution Approach 2:
The probabilistic module automatically learns and adapts to the specific characteristics of different call center applications through training on historical data. Instead of requiring manual tuning for each application, the system self-adjusts by learning from the data it processes, automatically optimizing its categorization logic for the specific domain without human intervention in the tuning process
3Ease of manufacture
If pre-defined binary rules are used for call categorization, then the setup process is straightforward, but it needs to be repeated in full for different call center applications
Solution Approach 1:
The probabilistic module is designed as a universal system that can be trained on historical data from any call center application, whether banking, retail, telecommunications, or other domains. The module learns application-specific patterns from the training data and adapts its categorization logic accordingly. This allows a single probabilistic model framework to serve multiple different call center applications without requiring complete reconfiguration, achieving both ease of setup and high adaptability
Solution Approach 2:
The system adapts to different call center applications by changing the parameters learned from training data rather than changing the fundamental system structure. The probabilistic module learns application-specific parameter values and relationships from historical data, allowing the same system architecture to effectively categorize calls across different domains by simply retraining with application-specific data
4Ease of operation
If pre-defined binary rules are used for call categorization, then the rule-based approach is easy to understand, but it does not work well for call reasons where there is no clear associated action from the agent
Solution Approach 1:
The probabilistic module serves as an intermediary between the simple binary rules and the complex task of categorizing calls without clear associated actions. For calls where binary rules are insufficient or ambiguous, the probabilistic module analyzes patterns in the data, considers multiple features simultaneously, and determines the most likely categorization based on learned relationships. This intermediary approach maintains the simplicity and interpretability of rule-based systems for clear cases while providing reliable categorization for complex cases through data-driven pattern recognition
Data Source
AI summary
A computing system is configured to categorize digital data records each relating to an interaction between a call center agent and a customer. The system includes a rules module to determine a first probability that a data record belongs to a category based one or more rules each rule defining a relationship between one or more features of the data record and the category; a probabilistic module to determine a second probability that the data record belongs to the category based on one or more probabilistic models of relationships between one or more of the features and the category; an arbitration module to determine whether or not the data record belongs to the category based on the first and second probabilities; and an action module to receive from the arbitration processor an indication that the interaction belongs to the category and to initiate an action appropriate to the category.