Privacy-Preserving Email Classification Using Metadata Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in effectively identifying and managing unwanted electronic communications, such as spam, phishing, and malware, within enterprise email systems while maintaining confidentiality and protecting Personally Identifiable Information (PII).
Innovation Solution
The development of a labeled training dataset for supervised machine learning models that classify emails without accessing confidential information or PII. This is achieved by using relationships between email features, such as metadata and transmission data, to determine appropriate labels for previously unlabeled emails, utilizing expansion graphs to capture inference logic, and employing a voting system to resolve conflicting labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If email content is accessed to identify unwanted communications, then classification accuracy is improved, but privacy protection deteriorates
Solution Approach 1:
The patent extracts only the metadata and transmission data from emails, separating these features from the confidential email content. By taking out only the necessary identifying features (sender address, timestamp, host server, URL hashes) while leaving the private content untouched, the system achieves classification accuracy without compromising privacy protection.
Solution Approach 2:
The patent segments the email identification process into two independent parts: (1) extracting non-PII features such as metadata and transmission data, and (2) using these segmented features for classification. This segmentation allows the system to work with only the necessary identifying characteristics while excluding sensitive content, thereby maintaining both accuracy and privacy.
2Measurement precision
If manual labeling of emails is performed to create training data, then label accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies preliminary automated clustering to group emails by their metadata features before manual labeling is needed. This preliminary action organizes the data structure in advance, so that when manual labeling occurs, it only needs to verify and refine pre-grouped clusters rather than labeling individual emails from scratch, significantly reducing processing time while maintaining accuracy.
Solution Approach 2:
The system implements feedback through iterative clustering and label refinement. Manual labels are applied to representative samples, then these labels feed back into the clustering algorithm to improve future automatic grouping. This feedback loop reduces the need for extensive manual labeling over time while maintaining high label accuracy.
Data Source
AI summary
Emails or other communications are labeled with a category label such as “spam” or “good” without using confidential or Personally Identifiable Information (PII). The category label is based on features of the emails such as metadata that do not contain PII. Graphs of inferred relationships between email features and category labels are used to assign labels to emails and to features of the emails. The labeled emails are used as a training dataset for training a machine learning model (“MLM”). The MLM identifies unwanted emails such as spam, bulk email, phishing email, and emails that contain malware.