Graph-Based Email Filter for Spam Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current electronic mail filtering techniques face challenges such as frequent updates needed for content-based filters due to spam adaptability, difficulty in obtaining reliable non-spam training data, and delays in non-content based approaches like blacklisting and authentication.
Innovation Solution
An electronic mail system utilizing a graph representation of email exchange history to assign an indicator of potential abusive mail character by analyzing sender and recipient relationships, leveraging social structure to classify emails without content analysis, and using non-spam emails classified with high confidence as training examples for statistical filters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If content-based filters are used to detect spam emails, then detection accuracy is improved, but frequent updates are required due to spam adaptability
Solution Approach 1:
The system performs preliminary classification of emails as non-spam with high confidence before they can be used for training. By proactively identifying and storing these emails in a training database, the system prepares training data in advance, reducing the need for frequent updates and allowing content-based filters to be refreshed less often while maintaining detection accuracy.
2Reliability
If manual classification is used to obtain reliable non-spam training emails, then data reliability is improved, but legal and privacy issues arise
Solution Approach 1:
The system enables self-service classification by automatically analyzing email characteristics and assigning non-spam labels with high confidence scores. This automated self-classification eliminates the need for manual human review, ensuring data reliability through consistent algorithmic criteria while avoiding legal and privacy complications associated with manual processing of user emails.
Solution Approach 2:
The patent replaces the mechanical process of manual email classification with an automated computational system. The filter component uses algorithmic analysis of email characteristics to objectively classify emails as non-spam, substituting human judgment with a reproducible mechanical process that avoids privacy intrusion while maintaining reliability.
3Ease of manufacture
If blacklisting or authentication methods are used for spam detection, then implementation simplicity is improved, but detection delay occurs
Solution Approach 1:
The system performs preliminary classification of incoming emails using the graph-based filter component before they reach the user. By pre-analyzing email characteristics and computing confidence scores in advance, the system reduces detection delay while maintaining implementation simplicity, as the heavy lifting is done before email delivery rather than during user interaction.
4Reliability
If participation of all email users is required for effective filtering, then filtering effectiveness is improved, but user burden increases
Solution Approach 1:
The system implements self-service filtering where the filter component automatically analyzes and classifies emails without requiring user action. Users simply receive their emails with spam already filtered, eliminating the burden of participation while maintaining effectiveness through automated analysis of email characteristics and graph-based relationship data.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The invention relates to a method of filtering an electronic mail in an electronic mail system and, an electronic mail system. The method comprises steps of: providing an electronic graph representation of electronic mail related characteristics derived from the history of electronic mail exchange in the electronic mail system in a data base, and assigning to a present electronic mail provided for transmission in the electronic mail system an electronic indicator indicative of a potential abusive mail character of the present electronic mail by analyzing present electronic mail characteristics based on the graph representation in the data base.