Communication Classification Platform for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing email protection techniques, such as blacklisting, are ineffective against sophisticated scams like 419 scams, which rely on social engineering and appear as legitimate messages, making it difficult for filters to distinguish between genuine and fraudulent communications, particularly affecting vulnerable users.
Innovation Solution
A communication classification platform that uses a combination of automated and human review processes, including a quick classifier, detailed classifier, and reputation system, to evaluate messages based on collections of terms, indicating terms, friendly email sources, and geolocation, assigning scores to determine the legitimacy of messages and filter out fraudulent ones.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If blacklisting techniques are used to filter email, then implementation is simple, but effectiveness against sophisticated scams is poor
Solution Approach 1:
The classification system is divided into multiple independent components: a quick classifier that performs initial filtering using simple blacklisting, and a detailed classifier that performs comprehensive analysis using multiple evaluation engines. This segmentation allows the system to maintain implementation simplicity at the first level while achieving high reliability through the second level.
Solution Approach 2:
The quick classifier acts as an intermediary between the incoming email and the detailed classifier. It performs preliminary filtering to eliminate obviously malicious messages before they reach the more complex detailed classification system, reducing the burden on the intermediary component while maintaining overall system effectiveness.
2Measurement precision
If comprehensive analysis is performed on all emails, then detection accuracy improves, but processing time increases
Solution Approach 1:
The classification process is segmented into two stages: quick classification for immediate filtering and detailed classification for comprehensive analysis. This segmentation enables the system to achieve high detection accuracy for suspicious messages while minimizing processing time for legitimate emails that are quickly identified and routed.
Solution Approach 2:
The system applies partial analysis (quick classification) to all emails and excessive action (detailed classification with multiple evaluation engines) only to messages that require further scrutiny. This approach optimizes the balance between processing speed and detection accuracy by avoiding comprehensive analysis of every email.
3Productivity
If automated filtering is used, then processing speed is high, but false positives increase
Solution Approach 1:
The system segments the filtering process into automated quick classification and semi-automated detailed classification with human review capabilities. This segmentation allows high-speed automated processing for clear cases while providing opportunities for human intervention to reduce false positives in ambiguous cases.
Solution Approach 2:
The system incorporates feedback mechanisms where classification results and user reports are used to continuously improve the classification algorithms and term collections. This feedback loop helps reduce false positives over time by learning from actual classification outcomes and user corrections.
Data Source
AI summary
Information associated with a plurality of electronic communications between a first entity and a second entity is obtained. First and second determinations are performed that the number of communications in the plurality of electronic communications exceeds a first threshold, and that the plurality of electronic communications were exchanged during a period of time that exceeds a second threshold. A classification is performed on a received electronic communication based at least in part on the first determination and the second determination, the electronic communication is assigned one of three different classifications: good, bad, and undetermined. At least one action is performed based at least in part on the performed classification.