twitter-tone.md

### Lab - twitter emtional tone

This time, you can build a fun, simple application that uses the Watson Tone Analyzer service.

You will need an instance of the Watson Tone Analyzer service to created from the IBM Cloud catalog. Switch to the IBM Cloud dashboard, select Catalog from the menu bar (top-right), and add " Tone" to the filter search argument.

Click on Tone Analyzer and create a new instance, accepting all defaults.

When the instance has been provisioned, you'll be presented with the service management panel -- select the Service credentials option from the left navigator. .

Then find and open the View credentials option:

Take a note of where the username and password fields are, as you will need to copy/paste these values into the Node-RED Tone Analyzer node, shortly.

The basic application flow is based on the following Cognitive Application Reference Architecture:

Let’s say we want to know about people’s general opinion on a specific topic on Twitter. We can use a twitter in node to obtain tweet feeds, then send the content to Watson Tone Analyzer service. When we get the analytic result, we will keep track of the result over time and visualize the average score of each index - similar to the following:

To produce the sample chart, you will need an additional package to be added to the Node-RED palette - node-red-dashboard.

From the Node-RED menu (top-right), select Manage palette.

Click on the Install tab and enter "node-red-dashboard" in the search modules field.

Click the install button.

All being well, you should see the additonal nodes being dynamically added to the list of available nodes:

You'll find these new nodes towards the bottom of the left-side palette menu.

Once more, create a new flow on the canvas with the following nodes:

The first time you use a twitter node, the application will need to be authorized to use Twitter (inbound and, optionally, outbound) using your Twitter credentials. This can be very simple if you have used Twitter through the same browser as you're using to access Node-RED:

1. in the twitter configuration panel, click the edit icon
2. when you see following prompt in a new tab or window, click Click here to authenicate with Twitter
3. you'll be asked to authorise the Node-RED application - click Authorize app
4. you should receive a message similar to Authorised - you can close this window and return to Node-RED -- close the tab or window, and return to the Node-RED editor view
5. click Add
6. click Done and you're good to go with Twitter integration.

For this Lab we will use #< your-event-hash-tag > as the Twitter topic to monitor.

Since this topic is likely to be VERY active, we have to avoid the flood of the incoming tweets overwhelming the Node-RED instance. Add a delay node after the twitter node and setting the node action to “Limit rate to” 1 message per 1 seconds - this should leave sufficient time for the Watson Tone Analyzer service to handle each request.

You will need to fill in the service credentials (username and password) from the Watson Tone Analyzer service that you created earlier.

In this example, the purpose is to extract emotion tones, so select the Emotion option in the Tones field.

Since the result from the Watson Tone Analyzer will be in msg.response, we can add a change node to move msg.response to msg.payload to make it easier to process in subsequent nodes.

In a function node named “Add tweet scores to total”, the scores can be accummulated using the following code:

var defaultResult = {
  "emotion_tone":{
    "Anger":  0,
    "Disgust":0,
    "Joy":    0,
    "Sadness":0,
    "Fear":   0
  },
  "count":  0
}

if(msg.payload) {

  var result = context.get('twitterAnalysis')||defaultResult;

  msg.payload.document_tone.tone_categories.forEach(function(toneCategory){
    if(toneCategory.tones){
      toneCategory.tones.forEach(function(tone){
        result[toneCategory.category_id][tone.tone_name] += tone.score;
      })
    }
  })
  result.count += 1;

  context.set('twitterAnalysis', result);

  return {payload:result};
}

Here, we add the scores of each tone and then save into a context variable named “twitterAnalysis”. A track of the count is kept, for the next node to use in calculating avergages.

The next function node calculates averages with this code:

if(msg.payload.count){

  msg.type = "newMsg";

  for (var toneCategory in msg.payload) {
    if (!msg.payload.hasOwnProperty(toneCategory)) continue;
    var obj = msg.payload[toneCategory];
    for (var prop in obj) {
      if(!obj.hasOwnProperty(prop)) continue;
      obj[prop] = obj[prop]/msg.payload.count;
    }
  }
}

return msg;

In order to show the scores of each tone, each score needs to be moved into different topics before sending to the dashboard chart node. We use the change node to do this; here is an example for the "Anger" score:

Note that the second Set section has been added using the + add button lower-left.

Modify the other change nodes in a similar way to reflect "Joy", "Sadness", "Disgust" and "Fear"

The chart node will need some configuration to allow it to display the rolling average of the 5 different emotion categories.

Click the edit button to establish a default dashboard for the chart.

and click Add.

Make sure the Legend option is updated to Show so you can distinguish between the various lines.

Once you redeploy the application, following the link to the Dashboard

and you should see the chart beginning to build:

Congratulations! - you have completed the process of integrating live Twitter with Watson's natural language processing capabilities for identifying emtional tone.