Understanding and managing machine learning in ZeroTime™
ZeroTimeTM uses machine learning to make associations between user activity and timesheet labels. It uses this information to make recommendations for label selections, with the goal of making time capture more efficient, accurate, fast, and easy.
ZeroTime makes its suggestions based on learned associations between timesheet labels and other input from entries, such as titles and comments. These are ‘soft’ suggestions, meaning the user can always choose to reject them.
Machine learning operates on a feedback loop, whereby the user’s denial or acceptance of suggestions influences future suggestions.
Machine learning data is not retained forever, and is continuously cleaned out over time, so the app will eventually forget associations if they’re not in use.
The machine learning process
This is an outline of the machine learning process in ZeroTime:
- User adds timesheet labels (like client, project, task or location) to notes, chatbot entries, and collector-mediated entries. They may include a # to help train the system.
- ZeroTime learns to associate timesheet labels with the content of the entry from the note and comment fields, and with other labels.
- Based on such criteria as correlation frequency, accuracy (based on feedback), and number of entries, ZeroTime reaches a confidence threshold for associations that triggers its recommendation engine.
- When the user adds notes or collectors generate entries, ZeroTime begins suggesting timesheet labels based its learned patterns.
- User gives feedback to the system by confirming or denying suggestions.
- Based on user feedback, ZeroTime updates its associations and improves the accuracy of its recommendations.
How long does it take for ZeroTime to learn a new association?
This depends. The recommendation engine doesn’t start making suggestions until a certain threshold level of associations is detected – that is, you have to associate two terms a certain number of times within a certain timeframe.