Sender Similarity Estimation for Bulk Message Reclassification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic message categorization systems are inefficient due to reliance on inaccurate classifiers and user distrust in automatic foldering, leading to manual re-categorization challenges, especially for users with large message volumes or limited interface functionality.
Innovation Solution
Implement a method to identify and correct categories for multiple messages simultaneously by determining similarity of message senders using co-occurrence matrices and pointwise mutual information, allowing for automatic recommendation and action on similar senders based on user actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic foldering is implemented to classify messages, then message categorization efficiency is improved, but classification accuracy deteriorates due to inherent limitations of classifiers
Solution Approach 1:
The system implements feedback by detecting user corrections to automatic classifications and using these corrections to identify similar messages. When users manually re-categorize messages, the system learns from this feedback and applies the same categorization to similar messages, thereby improving accuracy while maintaining automated efficiency.
Solution Approach 2:
The patent introduces an intermediary approach by using message similarity as a bridge between automatic classification and manual correction. Instead of relying solely on classifier accuracy or pure manual sorting, the system uses similarity metrics to transfer categorization decisions, mediating between automated and manual processes to achieve both efficiency and accuracy.
2Measurement precision
If manual re-categorization is performed to improve classification accuracy, then classification precision is improved, but time consumption increases significantly
Solution Approach 1:
The system performs preliminary action by pre-computing message similarity metrics and organizing messages into similarity groups before user interaction. When users correct one message's category, the system has already prepared the similarity information needed to quickly apply the same correction to related messages, reducing the time required for manual re-categorization.
Solution Approach 2:
The patent segments the large task of manual re-categorization into smaller, manageable units by grouping similar messages together. Instead of requiring users to review and re-categorize each message individually, the system presents groups of similar messages that can be re-categorized collectively, significantly reducing the time investment required.
3Measurement precision
If users review and correct each message individually to ensure accurate categorization, then classification accuracy is improved, but productivity deteriorates due to large message volumes
Solution Approach 1:
The system merges multiple individual message correction tasks into a single batch operation. By identifying similar messages and allowing users to correct them collectively, the system combines what would otherwise be separate productivity-draining tasks into efficient bulk operations, maintaining accuracy while preserving productivity.
Solution Approach 2:
The patent applies partial action by having users review and correct only a representative sample from each similarity group rather than every individual message. The system then extends this partial correction to the entire group, achieving comprehensive accuracy improvement with minimal user time investment.
4Device complexity
If simple automatic classification rules are used to maintain system simplicity, then device complexity is reduced, but adaptability deteriorates when handling diverse message types
Solution Approach 1:
The patent adds another dimension to the classification system by incorporating message similarity metrics beyond traditional classification rules. This additional dimension allows the system to handle diverse message types adaptively while maintaining the simplicity of the core classification engine, as the similarity computation operates independently from the classification logic.
Data Source
AI summary
A method assigns categories to electronic messages. Users assign labels to messages received from multiple senders. The user-assigned labels are aggregated into a co-occurrence matrix whose axes correspond to the plurality of senders. Each matrix entry measures the extent to which users have assigned identical user-assigned labels to the pair of senders. Using the co-occurrence matrix, a latent vector is computed for each sender. A user re-categorizes a message received from a first sender, assigning the message to a first category. One or more second senders are identified that are similar to the first sender. Similarity of senders is measured by proximity of the latent vectors corresponding to the senders. One or more second messages are identified from the second senders, and the user is presented with the suggestion to reclassify the second messages under the first category. Upon user confirmation, the second messages are reclassified.