...
New Feed Dataloader: RSS Feed Dataloader was reworked providing now more metadata from the RSS feeds to be mapped to facets and removing an issue where duplication of items happened.
Allow iframe embedding of mobile dashboards.
Make ground truth item endpoint sub-item ready. (Limit of 10 sub-items for returning all sub-items. If there are more sub-items, only the highlighted pages will be returned).
Powerful Transformers models: By upgrading the transformers library to 4.6.1, we now leverage transformer pipelines and use high-performing pretrained sentiment models to extract better sentiment results from text data.
For sentiment analysis, there is now an option to switch between using transformer-based pretrained models like ´finbert´ and ´distilbert´. For details about how to use it, follow the Squirro lib/nlp documentation here.
Improve toast duration for subscribing and unsubscribing communities from 4s to 2s.
Expose community ID to newsletters.
Apply highlighting to titles in all item details.
Added "label status" filtering to the Ground Truth labeling list and focus view.
Introducing a new flag
get_body_by_idto directly add the body of an item by submitting the item_id for document-level ground truth labeling.Expose community query to the JINJA template of the newsletters.
Introduction of additional filter options in ground truth labeling process, which allow the user to see all the labeled or not yet labeled elements of a candidate set.
Enable PDF viewer while labeling document-level ground truths.
Add a direct link to communities in newsletters.
Add the ability to apply document-level AI models on PDF documents.
Release 3.3.3 introduced improved performance/throughput of the Pipelet-Step through parallelised mini-batching. That change broke the capability of pipelets to work with projects having numeric or datetime facets. This has been fixed now in this release.
More stable Item Level Access Control: merge template query to parsed query-tree instead of modifying user query string.
...