3.4.5 Release Notes#

Features#

Frontend

[ML Workflow Studio Plugin] Section Query Processing: Those workflows got a new button to select a workflow to be used for query-processing.
Added tooltips for the Engagement map widget. Added drilldown on facet click.
[ML Workflow Studio Plugin] Added column for workflow type.
Many items widget and React improvements and fixes.
[ML Workflow Studio Plugin] Added sections in the studio plugin list for ML Workflows.
The standalone item detail view now has a close button to allow the user to navigate back to the dashboard space.
[ML Workflow Studio Plugin] Changed code editor styling when used with a studio plugin.
Query logging should use the original query if present.
Migrated Tabs widget to React.
New percentage mode for Metrics widget.
[New Items Widget] Added entities rows inside a card for sales insights app and config props to reflect that.

libNLP

Implemented decorator to cache selected steps.
Implemented cache in libNLP library.
Improve the optimization for the execution of multiple ML workflows by modifying the order of some libNLP steps.
Implemented caching mechanism for NLP ML workflows results.
Added field to the Document allowing to skip caching steps/workflows.
Added multi-lingual spaCyNormalizer: Perform language-specific text analysis that can then be used in subsequent steps.

Sales Insights

Added the ability to configure the facet containing the account name for the Insights Generator pipelet.
Added a signal-based generator class that generates insights cards per signal for different accounts.

Data Ingestion

The ingester service is now able to work with prioritized batches. This feature is controlled by the ingester.priorities.* options which are exposed in the Configuration panel available under the Server space.
The dataloader for parsing Squirro activity logs now works on multi-node Squirro deployments.

Query Processing:
- Please refer to Query Processing for information on query processing. You can also update the default workflow with the latest features directly from libNLP.
- Newly created projects have now a default query-processing workflow enabled (can be disabled via Project-Configuration). The default Query Pipeline Workflow gets installed as a global asset.
- Automatically set up the imported query-processing workflow in project’s configuration while importing a project template that comes with a configured query-processing ML workflow.
- Query-Processing-Pipeline Example: Query-Disambiguation through 0-shot classification
- Allow execution of QueryProcessor ML-Workflow in local (everytime /query endpoint is called) or global mode (only once triggered by the Global Searchbar via NLP /parse endpoint).
Initial implementation of question/answering custom widget (available on Cognitive Search template).
New Server Namespace Setting topic.monitoring.default-monitoring-project: setup default monitoring-project after first admin-user registration (per tenant).

Squirro Studio plugins

Upgrade instance:
- Added the functionality to allow the selection of channel and Squirro version.
Query Understanding:
- Reworked NLP-Parser plugin to run query-parsing with project’s configured query-processing ML-workflow. Moved to section “AI-Studio > Query Understanding”.
- Question-Answering endpoint: Search for a descriptive answer to a user’s question within top rated paragraphs of the results.

Frontend

The fixed text highlights not working for not labeled items in AI Studio.
Fixed text highlights scroll not working in AI studio.
Fixed <squirro> tags are sometimes being shown in the widgets that support highlighting.
Cards widgets now resize if they are configured to not display the abstract.
Fixed Chips widget text/action misalignment.
The spellchecking is applied to the result query instead of the original one.
Fixed an issue when creating a new custom widget, then right after a new dashboard, caused the dashboard to be unusable.
Fixed entity scrolling for PDF items.

AI Studio

Made NLP Runner work outside the ML Service context.
Pass error messages generated in the endpoints during HTTP requests back to the UI.
Ensure that the global_id is 22 characters long.

libNLP

Fix BERT sentiment token max limitation bug.
spaCy-Normalizer’s tokenizer is now able to consider phrases correctly (merges phrase tokens).

Data Ingestion

Improve MLEnrichmentGroupStep with type error on JSON serialization fix and ‘list’ object is not a mapping error fix.
Batches created from the Change Pipeline step will be of lower priority, letting newer batches to get executed first.
Do not group consecutive ML model steps into one when the machinelearning.optimize-workflows option is set to false.

Data Sources

Fix issue where changing the configured workflow of a data source was failing if its plugin could not connect to the source.
Fix and improve the validation of the provided pipeline steps configuration when creating a new or modifying an existing pipeline workflow.

Platform

Fix issue where entities created for PDF documents could lead to a mapping explosion.
Fix NLP-Tagger failing arbitrarily and blocking plumber process.
Fix pdf_table_extraction Studio Plugin by updating the version of the shipped camelot-py package to be compatible with the shipped pdfminer package.

Security

Update python-levenshtein package to prevent possible vulnerabilities.
Update tika-server to 1.28 (security update related to log4j https://tika.apache.org/1.28/index.html).

Squirro Studio plugins

Misc

Reduce log verbosity of PDF processing, making it more visible what is actually going on.
Migrate user preferences (user.user-store and user.json-preferences) from the old configuration service to the new one.

Rename “Natural-Language-Query-Parser” to “Query Understanding” and move it to the AI Studio section.

You will have to resolve at least the following config files when upgrading from Squirro 3.3.0:

For new installations, please follow the Installing Squirro on Linux instructions.

To upgrade an existing installation, please consult Upgrading Squirro.