Multi-factor Message Classification Using Segmented CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current email classification methods struggle with efficiently and accurately distinguishing between human-generated and machine-generated messages, particularly due to issues like 'sender cold start' and the lack of utilization of advanced natural language processing techniques.
Innovation Solution
A novel messaging framework that employs deep learning models at the message level, using four individual convolutional neural networks (CNNs) to analyze content, sender, action, and salutation patterns, and combines these models for a final classification that determines whether a message is human or machine generated, and categorizes it into specific MAGMA classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional logistic regression models are used for message classification, then the model structure is simple, but the accuracy and adaptability deteriorate due to sender cold start issues and lack of advanced NLP techniques
Solution Approach 1:
The patent segments the classification task into multiple independent CNN models, each analyzing specific message aspects (content, sender, action, salutation). This segmentation allows each model to specialize in one dimension, improving overall accuracy while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The patent creates a composite classification system by combining four different CNN models that process different message features. The composite approach integrates multiple perspectives (content analysis, sender analysis, action patterns, salutation detection) to achieve superior classification accuracy compared to single-model approaches.
2Measurement precision
If deep learning models with multiple CNNs are employed for message classification, then the classification accuracy improves, but the computational resources and model size increase
Solution Approach 1:
By dividing the classification task into four separate CNN models that each handle specific message aspects, the patent reduces computational burden compared to a single monolithic deep learning model. Each CNN can be optimized for its specific task, potentially using smaller architectures that collectively achieve the desired accuracy with reduced resource consumption.
3Reliability
If sender-level feature aggregation is used for classification, then the model can leverage historical behavioral features, but new senders with limited history cannot be reliably classified
Solution Approach 1:
The patent segments the classification approach by creating a dedicated sender CNN that analyzes sender-specific features independently from content analysis. This allows the system to handle new senders through content and pattern analysis even when historical data is limited, while maintaining reliable classification for established senders through learned behavioral patterns.
Solution Approach 2:
The patent changes the approach from aggregating features at the sender level to analyzing features at the message level through multiple CNNs. This parameter change enables the system to adapt to new senders by analyzing individual message characteristics rather than relying on accumulated sender history, improving adaptability while maintaining reliability through multi-factor analysis.
Data Source
AI summary
Disclosed are systems and methods for improving interactions with and between computers in content providing, searching and/or hosting systems supported by or configured with devices, servers and/or platforms. The disclosed systems and methods provide a novel messaging framework that automatically applies a multi-factor analysis technique to incoming and received messages in order to properly identify a message's type and category, which dictates the manner in which the message is displayed within a recipient's inbox. The disclosed framework operates on two levels: i) it determines whether a message is from a human or machine sender (H/M classification), and ii) it determines the messages category (MAGMA categorization).