Machine Learning Platform - Overview
The Machine Learning Platform (MLP) generates the training recommendations provided to users in various pages of the Cornerstone system. This platform applies a variety of analytical and machine learning techniques to user data, users' engagement and interaction with training, users' career mobility, learning trends, and learning over time. This collection of techniques is called the "Cluster Compute." The MLP applies many analytical techniques to generate recommendations for a specific user, based on their own portal.
Machine Learning provides the following benefits:
- Anticipates what a user might be interested in before they do
- Reduces workload for learning and development departments
- Shows trending content to users based on their associations
The Cornerstone MLP is a separate application from other Cornerstone product modules, such as Learning and Performance. For this reason, an integration was created to connect the MLP to applicable Cornerstone features, such as the Inspired by Your Subjects carousel and the Recommended for [User] carousel. Each feature correlates to a unique API which originates from the Cornerstone MLP.
See the following additional help topics for more information about the Machine Learning Platform:
- See Machine Learning Platform Features.
- See Machine Learning Platform Exclusions.
- See Recommendations Based on OU - Machine Learning.
- See Recommendations Based on User - Machine Learning.
Looking for an online course about how our Machine Learning Platform works? Click here: Machine Learning in the Learning Experience Platform
Below are answers to frequently asked questions:
What is requested and returned in each API call?
In each call, the application typically sends a request that includes the following:
- User ID
- Portal ID
- The number of items to return
When Sarah accesses the Inspired by Your Subjects carousel on the Learner Home page, Cornerstone Learning calls the API with Sarah’s User ID, her portal (Acme), and 500 training items.
The Cornerstone MLP then returns 500 training items in the API call (if enough recommendations are generated by the algorithmic techniques). Cornerstone Learning applies exclusions to the training items (e.g. ensures availability is respected), and displays up to twenty training items in the carousel on the Learner Home. See Machine Learning Platform Exclusions.
Why request 500 items and only display 20?
500 training recommendations are requested in order to ensure that at least twenty training items are available to appear in the carousel after all exclusions are applied . For example: The system requests 500 training items, and receives 150 MLP recommendations. Exclusions are applied to those results, which leaves thirty available results which can be displayed in a carousel. Twenty of these results can display in one of the user's carousels on Learner Home.
Why can't I report on recommendations?
Machine learning features are not stored anywhere in Cornerstone Learning. Instead, an API is called every time the user loads the page where the feature is displayed. For this reason, there is no way to pull a report of the training that is recommended to each user.
What is the Cluster Compute?
The Cornerstone MLP applies various algorithmic techniques to training data to generate training recommendations for users. This process is extremely complex and requires considerable computational power to run the algorithms for every user in every portal. This process is called the "Cluster Compute."
Running the Cluster Compute on the Cornerstone MLP requires many servers and a large amount of hardware to handle the load of analyzed data. Cornerstone continues to invest in more hardware to ensure the load is high-performing. However, due to the resources required, Cornerstone MLP runs according to a different schedule than other Cornerstone features.
What happens in the Cluster Compute?
The Cornerstone MLP analyzes current data in each portal for all users in order to generate recommendations for each individual user, based on the specific system feature. Each Cluster Compute is important because it analyzes the most up-to-date training registration and completion data to generate relevant machine learning recommendations for users.
For example: 100 users completed a training item yesterday. This data is not included in the machine learning recommendations from the last Cluster Compute. The next Cluster Compute will find this data, analyze it, and may recommend that training item to a specific user.
How long does the Cluster Compute take to process?
The Cluster Compute does NOT run instantly and may take days to process its data. The reason for this length of time is due to the volume of data to analyze for many users, for many portals, and many portal types.
The Cornerstone MLP includes Production portals, Stage portals, and internal portals that Cornerstone uses for testing in its Cluster Compute. All of these portals contribute to the large load in the overall cluster, which is why it may take days to complete each cluster run. Cornerstone has thousands of individual portals, and the Cluster Compute considers training registration data for all users, both active and inactive. This results in an immense amount of data to analyze.
Which portals are included in the Cluster Compute?
The Cornerstone MLP Cluster Compute currently includes Production portals, Stage portals, and internal portals that Cornerstone uses for testing. Note: Currently, the MLP does not include Pilot portals in its Cluster Compute. This is to reduce the load on the overall cluster to ensure that the process does not take too much time to complete.
The Cornerstone MLP supports Stage portals during User Acceptance Testing (UAT) to ensure clients can adequately test that the recommendations are working prior to enabling in Production portals.
How frequently does the Cluster Compute run?
The frequency of the Cluster Compute is dependent on the type of portal. This is due to the existing large load on the overall cluster.
The following portal types run according to the following schedules:
- Production portals: Approximately every 2 weeks
- Stage portals: Approximately every 2 weeks during UAT periods (less frequently between UAT periods)
Note: These times are subject to change, and the Cornerstone MLP team is working to decrease the amount of time it currently takes to run each Cluster Compute so that the frequency can be increased.
How can I tell when the last Cluster Compute completed?
To access the Cluster Compute Refresh Date and Time for each portal, please contact Global Customer Support, or check the Machine Learning Preferences page for the last cluster computation cycle time and date.
Note: The date and time listed are unique to each individual portal.
What happens if a new user is added to the portal after the last Cluster Compute ended and before the next Cluster Compute started? Will they see recommendations?
The user will not see personalized recommendations until the Cornerstone MLP performs the next cluster compute and learns which primary OU the new user belongs to. During the period of time before the cluster compute (before MLP knows the new user’s OUs), recommendations are provided to the new user based on what is popular across all OUs in the portal. Once the user’s OUs are recognized by MLP during the cluster compute, even if the user has not yet completed any learning, the bootstrapping mechanism will analyze the primary OU and its neighbors, if necessary, in order to deliver relevant recommendations based on the training signatures of other members of the new user's OU or neighboring OU(s).