Warning Message Classification via Sampling and Balancing

The patent segments the set of all warning messages into different priority levels or categories based on their importance. By dividing the monolithic set of warnings into segmented groups (e.g., critical, important, low priority), the system can present only the most relevant messages to developers while maintaining complete detection capability in the background.

The patent applies different quality standards or filtering criteria to different subsets of warning messages. Instead of treating all messages uniformly, it assigns local quality attributes (priority levels, severity ratings) to each message type, allowing developers to focus on high-quality (high-priority) messages while less important ones are automatically filtered or deprioritized.

The patent introduces counterbalancing techniques to offset the dominance of the majority class (ignored messages) in the training data. This may involve weighting schemes where minority class samples (fix messages) are given higher importance, or synthetic generation of additional minority class samples to counterbalance the imbalance and achieve better classification accuracy.

The patent performs preliminary data preprocessing actions before training the classifier, including resampling, balancing, or augmenting the data set to address the imbalance issue in advance. This preliminary action ensures that the training data has appropriate class distribution, enabling the classifier to learn effective decision boundaries for both classes.

The patent implements self-service mechanisms where the system automatically prioritizes and filters warning messages using trained classifiers and learned patterns from historical data. Instead of requiring developer intervention for each message, the system serves itself by automatically assigning priorities, filtering low-value messages, and presenting only the most relevant warnings to developers, thereby reducing triage time while maintaining accuracy.

The patent incorporates feedback loops where the system learns from developer responses to previously prioritized messages. By analyzing which messages developers actually fix versus ignore, the system refines its classification model and prioritization algorithms over time, improving accuracy while maintaining automated operation, thus avoiding the need for manual triage.