Call Traffic Anomaly Mitigation Through Probabilistic Reputation Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing call traffic management systems face issues such as 'flapping' and deterministic blocking, leading to undesirable network effects and blackouts in downstream analytics, while neural networks provide unexplainable and confusing outputs.
Innovation Solution
Implement a system that determines call reputation scores through random simulation using severity values as probabilities, allowing dynamic throttling and retraining on-premises, while adapting feature values without retraining the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If deterministic blocking based on positive/negative telephone lists is used, then call filtering is simplified, but flapping occurs causing network effects and analytics blackouts
Solution Approach 1:
The patent transitions from static deterministic blocking to dynamic probabilistic blocking. Calls from numbers on the negative list are blocked with a probability determined by a scoring model rather than always blocked. This dynamic approach prevents flapping by allowing temporary fluctuations in blocking behavior that smooth out network effects while maintaining overall filtering effectiveness.
Solution Approach 2:
The patent changes the parameter of blocking decision from binary (blocked/allowed) to continuous (probability value). The scoring model outputs a score that determines the probability of blocking, allowing nuanced control over call filtering. This parameter change enables the system to adjust blocking intensity based on severity without causing abrupt network effects.
2Device complexity
If deterministic blocking based on telephone lists is used, then implementation is simple, but downstream analytics experience blackouts
Solution Approach 1:
Instead of fully blocking all calls from negative list numbers, the system applies partial blocking by probabilistically allowing some calls through. This partial action maintains enough call traffic to keep downstream analytics functional while still providing robust filtering. The scoring model determines the probability, allowing controlled information flow.
3Measurement precision
If neural networks are used for call scoring, then detection accuracy improves, but outputs become unexplainable and confusing
Solution Approach 1:
The patent introduces an intermediary scoring model between the neural network detection and the final blocking decision. The neural network processes complex patterns to generate scores, which then feed into a more interpretable probabilistic blocking mechanism. This intermediary layer maintains detection accuracy while improving explainability by providing a clear score-to-probability mapping.
Data Source
AI summary
Methods and apparatus for call traffic anomaly mitigation are described herein. One or more embodiments include receiving a scoring request including a telephone number associated with a telephone call at a scoring device from a call processing entity, receiving a violator list of telephone numbers and their corresponding severity values at the scoring device from an anomaly analyzer, determining a severity value associated with the telephone number by performing a lookup operation in the violator list, performing a random simulation using the severity value as a probability to determine an indicator value, and inputting the indicator value into a model to determine a call reputation score.