DISCO AI’s tag predictions help lawyers review documents faster and more accurately. This guide will cover:
- Understanding DISCO AI's tag predictions
- Setting up DISCO AI tag predictions
- Using filters, search visualization, custom columns, and search to find DISCO AI tag predictions
- Using DISCO AI tag predictions while reviewing documents
Understanding DISCO AI's tag predictions
DISCO AI scores
DISCO AI predicts the likelihood that a given document should or should not be assigned a tag based on how documents with similar characteristics have been tagged in the past. For each document and tag pair, DISCO AI assigns a score from -100 to 100, where -100 means it is highly unlikely that the document will receive that tag and 100 means the document is highly likely to receive that tag.
| Prediction | Prediction score |
| Highly likely | 80 to 100 |
| Likely | 20 to 100 |
| Neutral | 20 to -20 |
| Unlikely | -20 to -100 |
| Highly unlikely | -80 to -100 |
DISCO's AI uses a systematic approach to ongoing machine learning, which means that its predictions are continuously updated as more documents are tagged. To improve accuracy, DISCO AI uses language-specific models for English, German, French, Portuguese, Spanish, Chinese, Japanese, and Korean.
DISCO AI's numerical scores are also accompanied by a color code that mimics that of a traffic light; green being likely to highly likely, yellow being somewhat likely or neutral, and red being unlikely to highly unlikely that a certain tag should be applied to a document.
| Color | What it means | Example |
| Green |
High scores – Highly likely this tag should be applied to this document |
|
| Lime green/yellow |
Medium scores – Likely or neutral that a tag should be applied to a document |
|
| Orange or red |
Low scores – Unlikely to highly unlikely that a certain tag should be applied to a document |
|
NOTE: You may notice some of the color squares are solid and some are not. A solid color square means that tag has already been applied to the document, while a hollow square means that tag has not been applied to the document.
Setting up DISCO AI tag predictions
Specifying tags for predictions
To get started, in the DISCO main menu, click Tags. On the Tags page, toggle AI Predictions to ON for the tags for which you want to turn on predictions.
Note: It is okay if you already have applied a particular tag to documents before you enable it for predictions. Those tagging activities are not lost and will be used to make predictions once you enable the toggle.
Reviewing and tagging documents
Review documents as you normally would. As you apply or don’t apply tags to documents, the DISCO AI learns from those actions. Once you have tagged 50 documents, DISCO AI will start to analyze those documents. When you tag 100 more (150 total) DISCO will start to show you recommendations on which documents should or should not be given the Broadband tag and the strength of that recommendation.
Tag predictions are reflected as a score ranging from -100 to 100 applied to every document in your database. A positive number indicates the likelihood a tag should be applied to a document, while a negative number indicates the likelihood a tag should not be applied to a document. This process of learning, training, and scoring documents based on your normal tagging activities happens continuously in the background.
Using DISCO AI tag predictions in search
Search Filters
Predicted Tags
You can filter for documents based on the likelihood that they will be assigned a specific tag in the Predicted Tags filters pane. In the Predicted Tags filters pane, you can select one or more score ranges for each tag of interest. The range begins with Highly likely and ends with Highly unlikely. Likely, Neutral, and Unlikely scores represent selections in the middle of the range. These scores indicate the likelihood a tag should be applied to a document.
The chart below shows how the DISCO AI predicted score ranges correspond to each likeliness range. For example, the prediction score range Highly likely uses 80 to 100 and Likely has a prediction score range from 20 to 100.
| Prediction score range | Equivalent search syntax |
| Highly likely | 80 to 100 |
| Likely | 20 to 100 |
| Neutral | 20 to -20 |
| Unlikely | -20 to -100 |
| Highly unlikely | -80 to -100 |
Predicted Tag Changes
You can filter for documents that DISCO AI predicts will either have a tag added or removed.
The filter includes a confidence range from Low to High. A High confidence setting means the predicted tagging system is highly confident that the documents returned are likely to receive tag changes, where as a Low confidence setting means the predicted tagging system is less confident that the documents returned will receive tag changes. DISCO returns a larger set of documents when a low confidence setting is selected.
Search Visualization
In the Predicted Tags visualization pane, hover over a colored section of the bar to see the documents associated with the section and the percent of the bar length that it represents. In the Predicted Tag Changes visualization pane, click one or more of the bars for tags of interest to filter results by whether tags are predicted to be removed or added to a set of documents.
For more information on how to use search visualization and filters in DISCO, see:
Search visualization quick start guide
Search columns
Using predicted tag filters with a custom view can help you quickly view and sort documents using predicted scores. Add columns from the specific tag predictions group to view individual predicted scores while using predicted tag filters. In this example, the view includes Broadband and Enron - Oil, Gas or Energy tag predictions.
Search syntax & search builder
The predicted tag score ranges are reflected below.
| Prediction score range | Search example for tag Broadband |
|
Highly likely (80 to 100) |
prediction(Broadband, Highly likely) |
| Likely (20 to 100) | prediction(Broadband, Likely) |
| Neutral (-20 to 20) | prediction(Broadband, -19 to -19) |
| Unlikely (-100 to -20) | prediction(Broadband, -79 to -20) |
| Highly unlikely (-100 to -80) | prediction(Broadband, -100 to -80) |
Using Search Builder can help you quickly view and sort documents using predicted score ranges. In Search Builder, build the search syntax using the Tag prediction and Tag buttons. Examples of search syntax for confidence settings are shown below.
| Confidence | Predicted to be added | Predicted to be removed |
| High | prediction(Broadband, 80 to 100) % tag(Broadband) | tag(Broadband) & prediction(Broadband, -100 to -80) |
| Medium | prediction(Broadband, 60 to 100) % tag(Broadband) | tag(Broadband) & prediction(Broadband, -100 to -60) |
| Low | prediction(Broadband, 20 to 100) % tag(Broadband) | tag(Broadband) & prediction(Broadband, -100 to -20) |
Document Viewer
When you are viewing a document in the document viewer, you can see recommended tags as well as tag predictions. By default, predictions are not displayed when viewing a document. To view tag predictions, in the TAGS section, click Show predictions.
The recommended tags are tags that have been enabled for predictions and have a score higher than 50 for that document, indicating a higher likelihood of being applied to a particular document.
To apply a recommended tag to the document, in the PREDICTIONS section, click the recommended tag to apply.
NOTE: DISCO AI does not apply any tags to documents automatically. Instead, it recommends tags to you to help reduce errors and find relevant documents faster.
You can also see the predictive score and color for all the tags predictive was turned on for when you click into the apply tag area.
From the prediction scores and colors displayed, you are able to see DISCO's AI predict that the Broadband and Enron Outside USA tags are relevant to this document, while the other three tags show very low relevance. You are able to confirm DISCO's predictions are correct by reading the document these predictions were coded for.
Review stage prioritization in workflow
You can leverage tag predictions to power a prioritized review in workflow review stages. You can configure your review stage to order the documents based on the predictions for a particular tag when they are batched for review. This could allow you to find the more relevant documents earlier in your review than if you performed a linear review. For a more detailed explanation on prioritized reviews with tag predictions, see Review Stage Management.
For a training video about DISCO's AI predictive workflow feature, see AI predictive workflow training.