Figure 4.

System architecture for community voting and database-driven classification. A PepBank user enters a vote through the web interface or a PepBank curator uploads votes in batch. Individual votes are aggregated and training labels assigned according to the majority vote. A native database extension sends a notification to the classification daemon. A task is placed in the queue, and the category is scheduled for reclassification. Upon completion, the classification results are read into the database and stored as the new prediction values. They are seamlessly displayed to the user through the web interface. The entire process is independent of the number of votes cast per category and takes less than 5 minutes per category.