Processing Text for Sentiment Analysis

Understanding emotion and communication is possible with Watson’s Tone Analyzing service. This post is a step-by-step guide to creating a messaging bot that returns sentiment based on inputted text.

Technologies: (Node + Express + Watson)

These are the steps you can take to create your sentiment-analyzing bot:

1. Create a new messaging application.
2. Create IBM Cloud credentials.
3. Create an express/node server that serves a POST endpoint.
4. Add watson-developer-cloud to your project.
5. Point your endpoint to the messaging application’s event subscriptions.
6. Add your bot to a channel. And start communicating.

This is the configuration for the node/ express. This project uses a digital ocean droplet, pm2 and nginx reverse proxy for this endpoint.

const express = require(‘express’);
const bodyParser = require(‘body-parser’);
const request = require(‘request’);
const app = express();
app.use(bodyParser.json());
app.use(bodyParser.urlencoded({ extended: true }));
const server = app.listen(4000);

The POST event returns URL verification and handles event callbacks to the events api:

app.post('/event', (req, res) => {
  switch(req.body.type) {
    case 'url_verification':
      res.send(req.body.challenge);
      break;
    case 'event_callback':
       const text = req.body.event.text;
       const channel = req.body.event.channel;
       analyzeTone(channel, text.replace('<@BOT_ID_GOES_HERE>',''))
      break;
  }
});

A map was created for certain emotions into emojis:

const emotions = {
  anger : '😠',
  disgust : '🤢',
  fear : '😱',
  joy : '😄',
  sadness : '😞'
}

The score for the tone in the range of 0.5 to 1. A score greater than 0.75 indicates a high likelihood that the tone is perceived in the utterance.

The unique, non-localized identifier of the tone for the results. The service can return results for the following tone IDs: sad, frustrated, satisfied, excited, polite, impolite, and sympathetic. The service returns results only for tones whose scores meet a minimum threshold of 0.5.function.

The confidence threshold >0.55 for this particular case. You can learn more about tone and utterance scores here.

You would need to communicate with the application programming interface and send out a POST:

function postMessage(channel, message) {
  let options = {
    method: 'POST',
    url: 'POST+MESSAGE+URL',
    form: {
      token: 'AUTH_TOKEN_GOES_HERE',
      channel,
      text: message
    }
  };
  request(options, (error, response, body) => {
    console.log(response.body)
  })
}

Test and start communicating with your bot.

Example 1: Inputting a message indicating sadness.

Example 2: Sending an excerpt.

We can understand more about text processing with this endpoint.

Enjoy!