| Review protocol | A Word or PDF document that contains a list of review tags (e.g., Responsive, Privileged) and their definitions. |
| Search protocol | An Excel or CSV file that contains a list of search terms in dtSearch format. |
| Review batch | A list of documents for review. The batch can be static (with a fixed set of documents) or dynamic (documents are sampled continuously from active learning). |
| Review assignment | A document in a review batch that is assigned to a reviewer or a group of reviewers. |
| Dataset | A subset of the document corpus filtered by search criterion. A dataset can be used in active learning, model training, evaluations, and exports. |
| Model | A machine learning model that can be used to predict how likely a document is relevant to a category or tag. |
| Instruction | A sentence or a paragraph that describes the criteria for assigning a document with a certain category or tag. |
| Predictions | The outputs of a machine learning model, including a prediction score for each document. The scores range from 0 to 1, with 1 indicating the most relevant and 0 indicating the least relevant. |
| Metrics | Evaluation metrics for measuring the accuracy of a machine learning model. It also monitors the prediction statistics, e.g., what percentage of documents is predicted as relevant. |
| Annotations | The outputs of reviewed documents, including tags, predicted tags, comments, highlights, etc. |
| Tags | Tags assigned by human reviewers. They are defined in the tag settings. |
| Predicted Tags | Tags predicted by machine learning models. |
| Classification task | The task of assigning tag(s) to a document as a whole. |
| Extraction task | The task of annotating parts of the documents, e.g. a text span or a region. |