BI: Articulate Intelligence for Power BI – V1


Quick Start Guide and Starter Project for Arria NLG’s Add-in for Power BI Desktop

Add natural language (NLG) narratives beside visuals to instantly enhance the analysis of your entire dashboard.


This documentation is organized as follows:

Using Arria’s Articulate Intelligence Add-in
  1. Open the sample Power BI report and import Arria’s custom visual
  2. Add a narrative to describe the visuals
  3. Create a project in Arria NLG Studio
  4. Modify the starter project
  5. Publish the NLG Studio sample project
  6. Link your published NLG Studio project to your Power BI report
Using Mapping Scripts
Optimizing the Narrative to Handle Filters
Contextualizing the Narrative
Going Further

 

 

Using Arria’s Articulate Intelligence Add-in

This guide explains how to get and use Arria’s NLG Studio add-in for Power BI Desktop.

You’ll begin by downloading a sample Power BI report. Next, you’ll import Arria’s custom visual into the report. Using Arria’s custom visual, you’ll add a narrative widget to the report. Then, using NLG Studio, you’ll create a very simple NLG project, which you’ll publish to the cloud. Finally, you’ll link the published Studio project to the narrative widget to bring its NLG into the report.

By moving through the steps of creating a very simple NLG project, you’ll get a glimpse of the amazing things you can do when you extend Power BI with Arria’s NLG technology.

Prerequisites:
Audience:

BI analysts, BI developers

Open the sample Power BI report and import Arria’s custom visual

1. Download this sample Power BI report.
2. In Power BI Desktop, open the report.

3. In the Visualizations pane (you may need to click Visualizations to open the pane), click the ellipsis (Import a custom visual).
4. Select Import from file.
5. Import the .PBIVIZ file provided by Arria.
You should now see the Arria logo in your Visualizations pane.

Add a narrative to describe the visuals

1. Make sure to click off the charts. (If you have a chart selected during the next step, the chart will be overwritten.)
2. In the Visualizations pane, click the Arria add-in icon to add the narrative widget to your report. Move it and resize it to fit to the left of the charts.

3. With the narrative widget selected, from the Fields pane, select the required fields: Product, Profit, Sales, Segment, Year.

4. Hover your mouse over the narrative widget and click the gear icon.
This is the configuration window of the narrative. You should see the selected data as a JSON object.

Note: Click the Focus Mode icon if you want to get a better look at the JSON data.

 

 

Now that you’ve selected the data you’re working with in Power BI (the fields Product, Profit, Sales, Segment, Year), you’re ready to bring that specific set of data into NLG Studio.

Important: Before bringing the data set into NLG Studio, you need to make sure you’ll get all the data you want. Make sure to select at least one whole visual. If only certain data points are selected in a visual rather than the entire visual, the data you download will be incomplete. (Also, for this example, make sure that in the Year visual, All is selected rather than 2013 or 2014). If you have a specific data point selected in a visual, left-click on that data point to clear the selection.

5. Click the Download link to get the data set.

6. In the Open Browser dialog, click OK.

7. Check your machine’s Download folder for the file called data.json. In the next section, you’ll import this into the BI Starter project in NLG Studio.

Create a project in Arria NLG Studio

You are going to create a very simple sample project in NLG Studio in order to get a basic understanding of how you can use NLG Studio to augment your analytics. (For instructions that show how to create a more elaborate narrative, see the NLG Studio documentation.)

1. Download the BI Starter project for NLG Studio. The BI Starter project contains some helper functions that you’ll see as sub-scripts under the Main script in the Compose view. These helper functions are user-defined functions created specifically for working with BI data.
2. Check your machine’s Download folder for the file called BI Starter Project.json.
3. Log in to NLG Studio.
4. In NLG Studio, in your dashboard, click the Import a Project icon to import the BI Starter Project JSON file.

5. In the NLG Studio dashboard, click the starter project to open it.

Note: In the future, this project may become your starting point for all your BI projects created in NLG Studio. So if you like, you can rename the project to something generic. Or, you might want to give it a name that matches the name of the corresponding .PBIX file. To change the project’s name, just click in the name at the top left of the screen.

You’ll see the data that comes with the starter project. When you’re ready to modify this starter project, you’re going to replace this data with the data from your field selections in Power BI.

6. In the left navigation ribbon, click Compose to access the Compose editor.

In the editor, you’ll see the contents of the Main script, which is empty aside from a reference to a ReadMe sub-script. This ReadMe describes the BI data format used in Power BI projects, and it provides information about the helper functions in the project. The helper functions appear as a list of sub-scripts under the Main script. You can click on the sub-scripts to see their contents (the user-defined helper functions).

7. Make sure you’re in the Main script, then In the left navigation ribbon, click Preview. Read the ReadMe information.

Modify the starter project

What you want to do first is write a piece of narrative describing the total sales and total profit for the selected period. To generate the text for that, you’ll add some code in the Compose editor. But first, you need to import the data set that is specific to the fields you have selected in Power BI: Product, Profit, Sales, Segment, Year.

1. In the left navigation ribbon, click Data to go to the Data view.
2. At the top right of the screen, click the icon to import data directly and bring in the file called data.json from your Downloads folder.

3. Click Compose to go to the Compose view.
4. Click Main to begin editing the Main script.
5. Remove the text that is already there ([[ReadMe]]).
6. Copy/paste the following:

For the [[inflectNoun('year',len(getUniqueValues(data,'Year')))]] [[getUniqueValues(data,'Year')]], the total sales were [[formatCurrency(columnAggregate(data,'Sales','sum'),'USD','#.00','')]] and the total profit was [[formatCurrency(abs(columnAggregate(data,'Profit','sum')),'USD','#.00','')]].

7. In the left navigation ribbon, click Preview.
Your previewed output should look like this:

Now add another piece of narrative. You’ll add a sentence reporting on the sectors that contributed the most and least profit. As an exercise, you’ll create some in-script variables that you’ll use to replace some of your existing code.

8. In Main, replace the existing script with the following:

[[global.currency='USD';
years=getUniqueValues(data,'Year');
totalSales=columnAggregate(data,'Sales','sum');
totalProfitLoss=columnAggregate(data,'Profit','sum');
profitLossbySegments=groupBy(data,'Segment','Profit','sum');
topSegmentbyProfit=top(orderDesc(profitLossbySegments),1)[0];
if (currency !=null){For the [[inflectNoun('year',len(years))]] [[years]], the total sales were [[formatCurrency(totalSales,currency,'#.00','')]] and the total [[direction(totalProfitLoss ,'profit','loss' )]] was [[formatCurrency(abs(totalProfitLoss),currency,'#.00','')]]. The [[printKey(topSegmentbyProfit)]] segment reported the most [[direction(topSegmentbyProfit[1], 'profit','loss')]]: [[formatCurrency(abs(printValue(topSegmentbyProfit)),currency,'0.##','')]].}]]

9. In the left navigation ribbon, click Preview.
Your previewed output should look like this:

10. Now add another sentence to your narrative, this time reporting on the loss made in some segments. Compose this sentence in a new script. Below the list of sub-scripts, click Add Script.

11. At the bottom of the list of scripts, overwrite the NewScript name with the name DescribeContrast.

12. In this new sub-script, copy/paste the following:

#define DescribeContrast(category,currency)[[salesForSegment=columnAggregate(filterOn(data,'Segment','==',category[0]),'Sales','sum');
if (category[1]<0){Even though the [[category[0]]] segment posted [[formatCurrency(salesForSegment,currency,'0.##','')]] sales, it reported a loss of [[formatCurrency(abs(category[1]),currency,'0.##','')]].}]]

In order to call the new sub-script, you need a new variable that represents the segment with the lowest profit (or highest loss). You can create an in-script variable in Main to do this.

13. Go back to the Main script and overwrite what you have there with the following code (which contains the new variable bottomSegmentbyProfit and the call to the new sub-script DescribeContrast):

[[global.currency='USD';
years=getUniqueValues(data,'Year');
totalSales=columnAggregate(data,'Sales','sum');
totalProfitLoss=columnAggregate(data,'Profit','sum');
profitLossbySegments=groupBy(data,'Segment','Profit','sum');
topSegmentbyProfit=top(orderDesc(profitLossbySegments),1)[0];
bottomSegmentByProfit=top(orderAsc(profitLossbySegments),1)[0];
if (currency !=null){For the [[inflectNoun('year',len(years))]] [[years]], the total sales were [[formatCurrency(totalSales,currency,'#.00','')]] and the total [[direction(totalProfitLoss ,'profit','loss' )]] was [[formatCurrency(abs(totalProfitLoss),currency,'#.00','')]]. The [[printKey(topSegmentbyProfit)]] segment reported the most [[direction(topSegmentbyProfit[1], 'profit','loss')]]: [[formatCurrency(abs(printValue(topSegmentbyProfit)),currency,'0.##','')]]. [[DescribeContrast(bottomSegmentByProfit,currency)]]}]]

14. In the left navigation ribbon, click Preview.
You should see the following:

Publish the NLG Studio sample project

1. In NLG Studio, in your BI Starter Project, click the Publish button at the top right of the screen.
2. At the Congratulations! message, make note of the URL for your project. (If you click inside the URL box, the URL is automatically copied to your clipboard. Paste it to a notepad.)
3. In the left navigation ribbon, click Settings to go to the Settings view.
4. In the Settings list at the left, click API, then Generate API Key.
5. Make note of the API key generated. (Again, if you click inside the box with the key, it is automatically copied to your clipboard. Paste it to a notepad.)

Link your published NLG Studio project to your Power BI report

1. In Power BI Desktop, in your report, click the Focus mode icon to maximize the narrative widget.

2. In the first empty field in the widget, enter the NLG service URL (the URL you made note of after publishing your NLG Studio project).
3. Add the authorization header.
a. In the Header field, enter Authorization.
b. In the Value field, paste in your API key, preceded by the word bearer as follows: bearer <your API key> 

4.  Click the Generate Text button (you may need to scroll down in the widget).

5. Click around on different pieces of the charts to see how the narrative responds. To select multiple bars at the same time, use CTRL+Click.

Using Mapping Scripts

With Arria’s Articulate Intelligence add-in, you can add custom JavaScript to modify the input data set — right inside the narrative widget. You do this in the Custom Mapping Script window.

About the Methods Used:

getData(): This method returns the underlying JSON data (the data you see in the Input Data in JSON Format window).
setData(appSpecificData): This method sets the appSpecificData variable (the transformed data object). After using this method, the transformed data object is available for download and posting to the NLG service URL.

The following example shows how you can add some metadata to the input JSON data object:
var data=getData();
var metadata={};
metadata["currency"]="GBP";
var dataset=data.dataset[0];
dataset["metadata"]=metadata;
var datasets=[];
datasets.push(dataset);
data["dataset"]=datasets;
setData(data);

To try out a custom mapping script:
1. Copy the above script and paste it in the Custom Mapping Script window of the narrative widget.

2. Click Download to get the sample data again, this time with your customization.
If you examine the downloaded data, you’ll see the additional information: ("metadata":{"currency":"GBP"})

3. In NLG Studio, open your BI Starter project.

4. Now assign the currency metadata to a variable called currency. You can do this easily as an in-script variable.
a. Replace the first line of the current Main script with the following:
[[global.currency=WholeJSON['dataset']['0']['metadata']['currency'];

To see the results:
1. In order to preview, you first need to import the modified sample data in to the Studio project. (This is the data you modified using the custom mapping JavaScript.)
a. In the left navigation ribbon, click Data to go to the Data view.
b. At the top right of the screen, click the Import Data icon.
c. Browse, or drag and drop, to import the modified data file from your Downloads folder.
d. Click Continue.
In the Data view, you should see the new "currency": "GBP" metadata.

2. Go to the Compose view, then from the Main script, click Preview.

3. Now publish the Studio project. The URL endpoint and the API key remain the same, so you don’t need to update anything more in the configuration window of the narrative widget in your Power BI report.
4. Go back to Power BI Desktop. Refresh the narrative (click Generate Text in the configuration window).

You should see the currencies expressed in sterling (£) instead of dollars (US$).

Optimizing the Narrative to Handle Filters

At this point, in your Power BI report, if you click on a single sector within a graph, the narrative still describes the sector with the most profit or loss. But this information is redundant because the selected data represents only a single sector.

To see an example of this redundancy, try this:
1. In the top graph, Ctrl-click to select all the bars in the Government segment like this:

2. Now check out how the narrative changes.

Do you see the second sentence? It’s reporting on the greatest profit by segment, which happens to be the Government segment. But because we’ve selected only the Government segment, we’re already getting the profit figure for that segment in the first sentence. We don’t need the second sentence. The next exercise will show you how to filter that out.

Note: By removing that second sentence, we are removing only one redundancy. The narrative could be further optimized. But we’ll leave that for another time.

So, focusing on one fix here, you can filter out that redundant sentence — the sentence describing the segment with the most profit or loss — when there is only one segment selected.

To filter out the sentence when it is irrelevant:
1. Locate the following sentence in your Main script:

The [[printKey(topSegmentbyProfit)]] segment reported the most [[direction(topSegmentbyProfit[1], 'profit','loss')]] of [[formatCurrency(printValue(topSegmentbyProfit),currency,'0.##')]].

2. Add the following if statement surrounding the sentence (include the curly brackets and replace <above script> with the sentence above):

[[if (len(profitLossbySegments)>1){ <above script> }]]

Note: If you preview the narrative at this point, you’ll notice that the sentence still appears in the narrative within NLG Studio. This is because Studio is working with the data it currently has, not the data that comes from the selection of only the Government segment in Power BI. To see that this redundant sentence has been removed, you need to re-publish the Studio project and then go back to Power BI.

4. Publish the Studio project again.
Now, in Power BI, you should see that the redundant sentence does not appear in the narrative when you select the Government segment.


Contextualizing the Narrative

To make the narrative more responsive to selection events within visuals, we’re going to add two prepositional phrases to the first sentence, so we can describe narrowed-down selections in segments and products. Currently, the first sentence states For the years <years>, the total sales were <sales total> and the total loss was <loss total>. To report on narrowed-down selections, we’re going to add these two new phrases after For the years <years>: in the segment <segment name> and for the product <product name>. An example sentence would look like this (new phrases in bold):

For the years 2013 and 2014, in the segment Enterprise, for the product Montana, the total sales were £2.61MM and the total loss was £31.10K.

1. First, create two new sub-scripts, one called SegmentPhrase and the other called ProductPhrase:

#define SegmentPhrase()[[
segment=getUniqueValues(data,'Segment');
if (len(segment)<=2) {in the [[inflectNoun('segment',len(getUniqueValues(data,'Segment')))]] [[getUniqueValues(data,'Segment')]],}
]]


#define ProductPhrase()[[
product=getUniqueValues(data,'Product');
if (len(product)<=2) {for the [[inflectNoun('product',len(getUniqueValues(data,'Product')))]] [[getUniqueValues(data,'Product')]],}
]]

2. In the Main script, find the year phrase ( [[if (currency!=null){ For the [[inflectNoun('year',len(years))]] [[years]],), and add calls to those two sub-scripts after the comma. The calls to the sub-scripts are shown in bold here:

[[if (currency!=null){ For the [[inflectNoun('year',len(years))]] [[years]], [[SegmentPhrase()]] [[ProductPhrase()]] the total sales were …

3. Preview the output. You won’t see the new phrases because they only show up with narrowed-down selections, but as long as you don’t get any errors, the code will work.

4. Publish the project, and try it out in Power BI Desktop. Below are examples of the results you should now see.

With all products in the Government segment selected:

With only the product Paseo selected, in the Government segment only:

 

Going Further

Familiarize yourself with the existing NLG Studio code and try out the Getting Started tutorials. Once you are confident, try to produce the following sentence to add to your starter project narrative. The added sentence describes the sectors having the highest and lowest profitability, for example:

The most profitable segment is Channel Partners (73%) and the lowest is Enterprise (-3%).

If you need help, here is the code for the new sub-script (Profitability):

#define Profitability() [[
sales=groupBy(data,'Segment','Sales','sum');
profit=groupBy(data,'Segment','Profit','sum');
profitabilityBySegments=sort(map(sales,x->(x[0], percentage(filter(profit,y->y[0]==x[0])[0][1],x[1],0,true))), (p,q)->p[1]<=q[1]?1:-1); highestProfitable=profitabilityBySegments[0]; lowestProfitable=profitabilityBySegments[len(profitabilityBySegments)-1]; if (len(profitabilityBySegments)>1){The most profitable segment is [[highestProfitable[0]]] ([[highestProfitable[1]]]%) and the least is [[lowestProfitable[0]]] ([[lowestProfitable[1]]]%).}]]

Here is the call to the new sub-script, which you can place at the end of the paragraph in your Main script:
[[Profitability()]]