Power BI Q&A Now Supports Customization of Natural Language Vocabulary

One of Microsoft’s Power BI features is the support for natural language based querying on Power Pivot models.  You can see a demo of this feature through a demo site that Microsoft has created using World Cup data.

Introduction video on Power BI Q&A

 

As I previously posted, the results are mixed – some phrases work well while others miss the mark.   As a cloud service, the results over time are expected to improve as Microsoft’s BING based machine learning algorithms optimize the query language.

Microsoft has now provided the ability to add your own “phrasings” to your data model using Office 365.  Microsoft calls this “Cloud Modeling” and you can optimize your model from the Power BI Site Settings Menu by adding your own custom phrasings and synonyms.

 

Optimize permissions are granted automatically to the person who uploaded the workbook.  In addition, they can be granted by the SharePoint Administrator.

For more detailed information on Q&A optimization, see this article.

The following are the types of custom vocabulary and grammar items you can add to your model:

  • Synonyms: add additional terms for tables, columns or field names in your model.  For example, if your table has a column called “Country” you might want to have a synonym for “Region” or “Location” depending on your context.
  • Phrases: add custom phrases based on nouns, adjectives, verbs and prepositions that resolve to queries in your data model.  These can be added to resolve ambiguities in user queries.

Power BI auto generates both of these based on your Power Pivot model when you upload your Excel workbook to SharePoint.   The customization feature allows you to add additional terms and grammar to help users search more effectively.

In addition to this customization, Power BI also provides a usage report that allows you to monitor how effective and usable queries are for your users.  Users have the ability to rate the search results and flag questions that didn’t match their expectations.  This will allow you to see how your tuning of the language models are improving the overall search effectiveness of your models.

Optimize Usage tab