Botminds V321 March 2018
Today we have released a new version - #V3– to live environment. This release adds four new features and also includes two critical bug fixes. In addition, as usual, we have included quite a few stability improvements and fixes. We will arrange separate training to our users towards enabling them to use these features efficiently.
New Labeling mode
With this release, a new mode of accessing segments with a focus on reviewing/labeling is introduced. You can enable this ‘ labeling mode’ from the options under your profile picture – ensure you have selected a project to get this option. In this mode, the annotation panel is shown first where you can see all the labels predicted/assigned for the selected section. You can do new or corrective annotations immediately which most of the time you can do with the information available in the segment. If you want to see the full document for more context you can load the document in the last pane.
With this release, you can use the ‘Ask Me’ panel to help you in reviewing the annotations done in the project. We are introducing a set of ‘commands’ for you to pull annotated segments for quicker review.
get all segments for a label /c /s ‘Skill Name’ /l ‘Label Name’
get all segments annotated by a user /u ‘User Name’ /a
get all segments approved by a user for a label /c /u ‘User Name’ /s ‘Skill name’ /l ‘Label Name’ /ap
get all segments annotated in a given time period /a /st ‘start time’ /et ‘end time’
get all segments approved or annotated by a user in a given time period /a /ap /st ‘start time’ /et ‘end time’ /u ‘User Name’
Get the complete list of commands by typing /help. You can use commands together appropriately.
Rapid Dataset generation through commands
With this release, creating the dataset for segment tagging made super easy. Search using some keywords or patterns and assign all of the sections to a label with a single command.
Example: /c Chennai /s ‘Address Extractor’ /l ‘City’ /assign
All the segments in the selected project that contain the word ‘Chennai’ will be assigned to the given label ‘City’ under skill ‘Address Extractor’
Further, we can fast track keyword extraction as well as using a similar command.
Example: /c Chennai /s ‘Address Extractor’ /l ‘City’ /extract
Rejecting a predicted label
With this release, we add a simple way to reject a prediction given by the trained model. In the annotation panel, for any predicted label, you can find a way to reject the suggestion. Rejections are getting logged for improving the model when the next iteration of model generation happens. Rejected labels will not be part of the context menu used to capture keywords.
- Conversion failures are properly marked as ingestion failures
- Duplication of extracted values in document summary reported in some scenarios got fixed