| Cold Start | When an item or user has just been added to your product and has minimal behavioral or context information. |
| Embedding | Compressed numerical representations that encapsulate the underlying features or attributes of the data type. |
| Model deployment | The process of creating or replacing a new model that can be accessed through the rank endpoint. Typically performed after a new model is trained. |
| Feature | An attribute associated with a user, item, or interaction that is used as input to your model. For example, age could be a user feature, price could be an item feature, device type could be an interaction context feature. |
| Interaction | An event that relates a user to an item. For example, a user replying to a social post, giving a movie a rating, a user following another user. |
| Item | Anything that a user interacts with. These are the things you’ll want to rank. For example: social posts, movies or other users. |
| Model | A machine-learning program that can make predictions from a given input based on patterns of previously seen data. |
| Online Store | A store for the features that are fed into a machine-learning model. This is a low-latency key-value store that typically contains the user and item context features. |
| Prediction | An estimation from a model. In our case this is typically a prediction of the most relevant items for each user. |
| Ranking | An optimal ordering of items surfaced to your users. This is what Shaped can find for you. |
| Recommendation | An optimal ordering of items surfaced to your users. At Shaped we keep this definition the same as ranking. |
| Training | When a model is training, it is looking at historical data and “learning” the relationships between your users and items. A trained model can be used to predict the items that are most relevant to a user (even if the user or item wasn’t seen at train time). |
| User | A product user identifier. For example, a logged in user, session, cookie or group identifier. |
