Overview of Pipeline Types
Machine Learning Pipelines are a way to automatically transform raw data into machine learning predictions. Cortex is a platform that allows anyone within an organization to easily create predictions based on data that has already been captured. Below, we will go over the different pipeline types available within Cortex, and list some examples to give a better understanding of what is possible.
Which Pipeline Should I Use?
Cortex offers multiple pipeline and prediction types. The following diagram will help explain which pipeline type is best suited for different predictions.
Future Events Pipelines
Describe an event in everyday terms, and Cortex will predict the likelihood that each user completes that event in the future. If you can describe it with your data, you can predict it in Cortex – any type of event, any set of conditions, and any future window of time. Learn how to build your own Future Events Pipeline.
Future Event Pipeline Examples
- What is the probability that each user purchases from category “shoes” within 7 days?
- What is the probability that each user will take any action within the next 30 days?
- What is the probability each Visitor to our site will Subscribe?
- What is the probability a Lapsed User will Resubscribe?
Look Alike Pipelines
Upload a list of users which share a common characteristic. Cortex will search through the rest of your users and score each one in terms of how similarly it looks and acts to the uploaded set. If you have a list of users without the characteristic, use a Classification pipeline instead. Learn how to build your own Look Alike Pipeline.
Look Alike Pipeline Examples
- How similar is each user to an uploaded set of homeowners?
- How similar is each user to an uploaded set of first-time homebuyers?
Upload a list of users which share a common characteristic, and another list of users that don’t exhibit the characteristic. Cortex will score each remaining user in terms of its likelihood to belong to the first group rather than the second.
Like Look Alike pipelines, Classification is used to determine whether your users are likely to belong to a certain group. If you have access to negative labels, you should always use a Classification pipeline to solve this type of problem, since negative labels give Cortex more information from which to learn. Learn how to build your own Classification Pipeline.
Classification Pipeline Examples
- What is the likelihood that each user is a homeowner, based on sets of homeowners and non- homeowners?
- What is the likelihood that each user is a student, based on sets of students and non-students?
Upload a list of users along with values of some numeric attribute. Cortex will predict the value of that attribute for each of your remaining users. Learn how to build your own Regression Pipeline.
Regression Pipeline Examples
- What is the current age of each user, based on an uploaded set of known user ages?
- What is the household income of each user, based on an uploaded set of known user incomes?
Recommendations within Cortex don’t require a specific Pipeline setup, as they are run automatically in the background. Recommendations are based on the Items (see Items API) ingested into Cortex, and uses the behaviors of all customers to predict the best recommended items for each user.
Onsite User Targeting
- Targeted Promotion: Promote commercial products onsite to those users most likely to purchase those products using the Predictions API call.
- Targeted Feedback: Solicit product feedback or collect information by targeting users likely to complete a pop-up survey with the Predictions API call.
Onsite Item Modules
- Recommended For: Display personalized content or item recommendations to each of your users with the User Recommendations API call.
- Similar Items: When a user engages with an item, show similar items suited to that user with the Personalized Item Similarities API call.
- Popular: Display the most popular items within your library or catalog with the Popularity API call.
- Trending Items: Display trending items (based on recent activity) within your library or catalog with the Trending API call.
- Recently Viewed: Show to each user a list of items that user has recently interact with using the Recent Items API call.
- Item Ordering: Display a set of items in the optimal order of preference for each user with the User Item Rankings API call.
- Item-Specific Push: Send item-specific push notifications to the users most likely to click on that item using the Top Users per Item API call. Effectively announce a new item or promote an existing one in a highly targeted fashion.
- Personalized Email: Send personalized emails with unique item recommendations for each user with the User Recommendations API call.
- What is a Machine Learning Pipeline?
- How to Build a Future Events Pipeline
- How to Build a Look Alike Pipeline
- How to Build a Classification Pipeline
- How to Build a Regression Pipeline
Still have questions? Reach out to firstname.lastname@example.org for more info!