FCRF Detection via Topic Modeling and Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting First Contact Resolution Failures (FCRF) in organizations are prone to precision errors and low recall due to their reliance on lexicons and pattern-based techniques, especially when dealing with noisy textual data from transcribed telephone calls.
Innovation Solution
An automated system using interaction topic modeling and model-based classification to detect and analyze FCRF, which includes pre-training a classification model with tagged pairs of interactions and utilizing parameters such as topic distance and external data like interaction date and customer identity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If lexicon-based information extraction methods are used to detect FCRF, then the detection process is simple and fast, but the precision and recall are low due to transcription errors and implicit statements
Solution Approach 1:
The patent introduces topic modeling as an intermediary layer between the noisy transcribed text and the final FCRF detection. The topic modeling process transforms the raw transcript into topic distributions, which serve as a robust representation that is less sensitive to transcription errors. This intermediary representation enables more accurate pattern matching and clue detection while maintaining computational efficiency.
Solution Approach 2:
The patent changes the parameter space from raw text matching to topic distribution comparison. By transforming the detection problem from lexical pattern matching to topic-based analysis, the system achieves higher precision and recall. The topic distributions capture the semantic meaning of interactions, making the detection robust to transcription variations and implicit statements.
2Measurement precision
If topic modeling is used to provide semantic depth, then the detection precision improves, but the computational complexity increases
Solution Approach 1:
The patent performs topic modeling as a preliminary step before FCRF detection. By pre-computing topic distributions for all interactions and storing them, the system avoids repeated complex computations during detection. The pre-processed topic representations are then efficiently queried and compared during the detection phase, reducing overall computational complexity while maintaining high precision.
Solution Approach 2:
The patent extracts only the essential topic distribution features from the full topic modeling output for FCRF detection. Rather than using all topic modeling results, the system selectively extracts the relevant topic proportions and comparisons needed for FCRF analysis. This extraction approach reduces computational complexity while preserving the semantic depth benefits.
Data Source
AI summary
The subject matter discloses a method for detection and analysis of first contact resolution failures comprising: obtaining a first plurality of digital representations, each digital representation of the first plurality is associated with a contact center interaction; determining a pair of digital representations that represent a pair of contact center interactions determined from the first plurality of digital representations according to metadata associated with the pair of contact center interactions; determining topics of each interaction of the determined pair of contact center interactions represented by the pair of digital representations; classifying the pair of contact center interactions as first contact resolution failure or first contact resolution interactions pair.