User permissions required: ‘View Sources’ AND ‘View Labels’
The Explore page allows you to search, review and filter a dataset to inspect and review individual verbatims.
By default, Explore presents the 20 most recent verbatims in a dataset in the 'Recent' mode, you can click the dropdown mode selector in the top left-hand corner of the page to change this.
Dropdown mode selector
The different options you can select from the dropdown menu are:
- Recent – view 20 most recent verbatims
- Shuffle – view 20 random verbatims
- Teach – show 20 verbatims that Re:infer is unsure how to label
- Low confidence – show 20 verbatims for which Re:infer currently has very low confidence about which labels may apply
- Rebalance - show 20 verbatims that are underrepresented by the training data in your dataset
- Label - view 20 verbatims with the selected label assigned / predicted (this is the default mode when a label is selected)
- Check label - view 20 verbatims that may have the selected label applied incorrectly
- Missed label - view 20 verbatims that may be missing the select label
At the bottom of the page you can click to move to the next page of 20 verbatims, or go back to a previous page.
To understand more about how to use Explore, see here.
Filter bar and adding or removing filters
The filter bar on the left-hand side of the page (as shown below) allows you to find specific groups of verbatims.
From this filter bar you can filter to:
- Specific date ranges (pick exact dates or select from options like the last week, month, 90 days or year)
- Reviewed or unreviewed verbatims
- Verbatims with positive or negative sentiment predictions (if sentiment is enabled on the dataset)
- Add any filter based on the metadata properties associated with your verbatims (click 'Add a new filter')
- Filter to verbatims that have specific entities predicted or assigned
- Filter to verbatims that have (or do not have) a specific label or combination of labels predicted (see the article on Advanced Prediction Filters for more detail)
Adding a user property filter
When you add a filter to metadata fields with a string format, you will be able to choose which to include or exclude in your selection (as shown in the two examples below):
Example inclusion and exclusion filters for a string format metadata fields
Please Note: When you choose to include certain values for a user property, you automatically exclude the remainder (and vice versa).
If you add a filter to metadata fields with a number format, you will be able to select minimum or maximum values (as shown below), to create a range of your choice:
Filter for number format metadata fields
Removing a user property filter
To remove a filter that you've applied, simply click the bin icon that appears when you hover over it with your mouse (as shown below), or select 'Clear All' at the top of the filter bar, to remove all filters applied.
Filter showing the 'remove filter' icon
Please Note: The introduction of Advanced Prediction Filters means that the label filter bar no longer automatically selects the 'Label' training mode, nor do you use it to select a label to train whilst in another mode such as 'Check label', this is now done via the taxonomy dropdown that appears after you select a training mode from the dropdown menu shown above.
You can use the label filter bar to filter to verbatims that have (or do not have) specific labels predicted, either whilst Model Training or whilst exploring and interpreting your data. To see how they work in more detail, see the article on Advanced Prediction Filters here.
You can use the buttons at the top of the label bar to filter between showing all verbatims, to those that have had labels assigned to them, or those with predictions (that have not been reviewed). The icons are shown below, and they change colour when selected:
|Select verbatims that have assigned labels|
|Select verbatims that have labels predicted|
To deselect the filter, simply click the button again.
If you select neither button, but filter to a label, the platform will filter to all verbatims that either have the label pinned or predicted, starting with the reviewed verbatims first.
Label bar guide
The label filter bar and '+ Add label filter' allow you to add complex combination or inclusion and exclusion filters (i.e. show me verbatims with X and Y predicted, but not Z). To find out more about how to use these, see the 'Advanced Prediction Filters' article here.
Red dial training indicator:
- The red dial training indicator (see here for explanation) next to some labels highlights those which require more training examples for Re:infer to accurately evaluate the performance of the label
- The completeness of the circle indicates how many more examples are needed. The larger the red section, the more examples are required
- Once you have 25 labelled examples, the red circle will disappear (depending on the complexity of the label, however, you may need more examples to get accurate predictions)
- You should review verbatims to find more training examples