Multi-pass Ad-hoc Email Spam Filter with Memory-Resident Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing message filtering technologies often misclassify messages as spam, leading to oversights where legitimate messages are filtered out, as they rely on generic classification models that do not account for individual user preferences and context, resulting in privacy concerns due to persistent storage of classification data.
Innovation Solution
A multi-pass message filtering approach that iterates through message folders, using a temporary, memory-resident classification model to retrospectively inspect messages and provide recommendations for reclassification, distinct from initial filters, ensuring privacy by maintaining data only in non-persistent memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If generic spam filters are used to filter messages, then filtering coverage is improved, but message classification accuracy deteriorates due to inability to account for individual user preferences
Solution Approach 1:
The patent segments the filtering process into multiple passes: a first pass using a generic classification model for broad filtering coverage, and a second pass using an ad-hoc classification model for individualized accuracy. This segmentation allows each pass to specialize in different aspects of filtering, resolving the contradiction between coverage and precision.
Solution Approach 2:
The patent performs preliminary filtering with a generic model before applying individualized ad-hoc filtering. The first pass pre-sorts messages into candidate sets, reducing the workload for the second pass and enabling efficient individualized processing without sacrificing overall filtering coverage.
2Measurement precision
If classification data is persistently stored for improved filtering performance, then filtering accuracy is improved, but user privacy deteriorates due to sensitive data storage
Solution Approach 1:
The patent uses temporary, non-persistent memory for storing classification data during the ad-hoc filtering process. The data is discarded after use, eliminating long-term privacy risks while maintaining filtering accuracy during the active filtering session. This disposable approach to data storage resolves the contradiction between accuracy and privacy.
Solution Approach 2:
The patent introduces temporary memory as an intermediary between the classification process and persistent storage. This intermediary allows accurate classification to occur without committing sensitive data to permanent storage, thus maintaining privacy while achieving filtering accuracy.
3Measurement precision
If multi-pass filtering is applied to correct misclassifications, then classification accuracy is improved, but processing time deteriorates
Solution Approach 1:
The patent applies ad-hoc filtering locally only to candidate message sets identified by the first pass, rather than processing all messages uniformly. This localized approach concentrates computational resources on messages that need reevaluation, improving accuracy while minimizing overall processing time.
Solution Approach 2:
The patent performs partial filtering by applying the computationally intensive ad-hoc model only to a subset of messages (candidate sets) rather than all messages. This partial action achieves the necessary accuracy improvement without the full time cost of universal reprocessing.
Data Source
AI summary
Embodiments are directed towards employing a multi-pass ad-hoc spam message filtering approach that dynamically generates a temporary classifier during a first pass based on a result of a previously applied message filter that sorts messages into various folders for a user. The first pass scans messages in a user's mail folders, and reads various information within the messages, including, but not limited to sender information, headers, including a subject, an originating network address, message contents, attachments, and the like. After creating a classification model, the classifier with its model is used in a second pass on the message folders to retrospectively inspect the messages and present to the user a list of messages that might be misclassified. The classification model is maintained within memory on a user's client device, as memory resident only, and is not stored on disk or within another persistent data store.