Adding Emotional Tone Analysis to Your Contact Form

A few days ago I blogged about adding customized form handling to your static site at Netlify. This was done via a simple serverless function that listened for form submissions and used the SendGrid API to send an email. While this works just fine, I actually had something more interesting in mind that I had to delay a bit. Imagine if instead of just getting emails about contact form submissions, you actually got something with a bit of a warning in terms of their content:

In the image above, you can see some basic information about the contents of the email based on their tone. This would be a great way to know what to prioritize in terms of reading and responding. To build this, I made use of the IBM Watson Tone Analyzer service. I’ve used this multiple times in the past with various serverless demos with OpenWhisk, but I thought I’d give it a shot with Netlify and Lambda. Here’s the full script (and be sure to read the last entry for context on how it works) with the new feature added:

const helper = require('sendgrid').mail;
const SG_KEY = process.env.SENDGRID;

const axios = require('axios');

const ToneAnalyzerV3 = require('watson-developer-cloud/tone-analyzer/v3')
const TA_KEY = process.env.TONEANALZYER;

exports.handler = async (event, context, callback) => {
    console.log('submission created error testing');

    let payload = JSON.parse(event.body).payload;
    let analysis = '';
    let toneString = '';

    //lets analyze the text

    if(payload.data.comments && payload.data.comments.length) {
        analysis = await analyze(payload.data.comments);

        /*
        if we get results, its an array of tones, ex:

        [ { score: 0.633327, tone_id: 'fear', tone_name: 'Fear' },
        { score: 0.84639, tone_id: 'tentative', tone_name: 'Tentative' } ]

        So what we will do is create an analysis string based on tone_names where score > 0.5
        */
        analysis = analysis.filter(t => t.score > 0.5);
        // and now we'll build an array of just tones
        let tones = analysis.map(t => t.tone_name);
        // and then a string
        toneString = tones.join(', ');
    } 

    // note - no validation - booooo
    let from_email = new helper.Email(payload.data.email);
    let to_email = new helper.Email('raymondcamden@gmail.com');
    let subject = 'Contact Form Submission';

    if(toneString.length > 0) subject += ` [Tone: ${toneString}]`;

    let date = new Date();
    let content = `
Form Submitted at ${date}
--------------------------------
`;

    for(let key in payload.data) {
        content += `
${key}: ${payload.data[key]}
`;
    }

    let mailContent = new helper.Content('text/plain', content);
    let mail = new helper.Mail(from_email, subject, to_email, mailContent);
    let sg = require('sendgrid')(SG_KEY);

    let request = sg.emptyRequest({
        method: 'POST',
        path: '/v3/mail/send',
        body: mail.toJSON()
    });

    await sg.API(request, function(error, response) {
        if(error) {
            console.log(error.response.body);
        } else console.log(response);
    });
    console.log('And done...');
};

async function analyze(str) {
    console.log('going to tone analzye '+str);

    let toneAnalyzer = new ToneAnalyzerV3({
        username: 'apikey',
        password: TA_KEY,
        version: '2017-09-21',
        url: 'https://gateway.watsonplatform.net/tone-analyzer/api/'
    });

    const result = await new Promise((resolve, reject) => {
        toneAnalyzer.tone(
            {
                tone_input: str,
                content_type: 'text/plain'
            },
            function(err, tone) {
                if (err) {
                    console.log(err);
                    reject(err);
                } else {
                    resolve(tone.document_tone.tones);
                }
            }
        );
    });
    return result;

}

Alright, let’s break this down. First, I load in the Watson Node.js SDK. While this isn’t necessary, I had issues using the REST API for Tone Analysis directly and decided to simply take the easy route out and use their package.

const ToneAnalyzerV3 = require('watson-developer-cloud/tone-analyzer/v3')
const TA_KEY = process.env.TONEANALZYER;

Where does the process.env.TONEANALZYER key come from? Don’t forget you can define custom environment variables for your Netlify sites.

Next, let’s see if we have data to check. In this case I’m assuming I’ve got a field called comments and it’s a block of text. You can make this more generic, or even use hidden form fields as a way of saying what should be checked.

let analysis = '';
let toneString = '';

//lets analyze the text

if(payload.data.comments && payload.data.comments.length) {
    analysis = await analyze(payload.data.comments);

Note the fancy use of await. As a warning, please note I’m still fumbling my way around async/await. Let’s look at analyze:

async function analyze(str) {
    console.log('going to tone analzye '+str);

    let toneAnalyzer = new ToneAnalyzerV3({
        username: 'apikey',
        password: TA_KEY,
        version: '2017-09-21',
        url: 'https://gateway.watsonplatform.net/tone-analyzer/api/'
    });

    const result = await new Promise((resolve, reject) => {
        toneAnalyzer.tone(
            {
                tone_input: str,
                content_type: 'text/plain'
            },
            function(err, tone) {
                if (err) {
                    console.log(err);
                    reject(err);
                } else {
                    resolve(tone.document_tone.tones);
                }
            }
        );
    });
    return result;

}

This basically just wraps the call to the Tone Analyzer API and returns the result data. I kept this mostly generic. Now back to the caller:

/*
if we get results, its an array of tones, ex:

[ { score: 0.633327, tone_id: 'fear', tone_name: 'Fear' },
{ score: 0.84639, tone_id: 'tentative', tone_name: 'Tentative' } ]

So what we will do is create an analysis string based on tone_names where score > 0.5
*/
analysis = analysis.filter(t => t.score > 0.5);
// and now we'll build an array of just tones
let tones = analysis.map(t => t.tone_name);
// and then a string
toneString = tones.join(', ');

As the comments say, you get an array of tones back and they do not appear to be sorted. I did a quick “quality” filter by removing tones with a score less than 0.5. That was arbitrary. I then map out just the name and finally make a string.

By the way, I’m 99% sure those three things could be done in one fancy line of JavaScript by someone who can work at Google. I don’t work at Google.

The final bit is to simply add the tones if we got em:

if(toneString.length > 0) subject += ` [Tone: ${toneString}]`;

And that’s it! So let’s have some fun with this. Warning, adult language incoming. If the adult language doesn’t make sense to you, ask your kids.

I AM SO FUCKING MAD AT YOU I WISH YOU WOULD DIE I HATE YOUR SERVICE.
I HATE EVERYTHING YOU DO.
I HATE KITTENS.
i HATE PUPPIES.
I HATE BEER.

OH MY GOD IM SO MAD AT EVERYTHING

This returned what you would expect: Contact Form Submission [Tone: Anger]

Now check this input:

i'm so happy with your service, but i'm nervous that if i commit to a monthly
payment i'll not actually make use of it enough to get value. can you give me
some more details on what i get with this service and help convince me it's worth
it?

While I know Watson isn’t perfect, but wow, check the result: Contact Form Submission [Tone: Tentative] I’d consider that near perfect.

You could imagine connecting this with some rules in your mail server such that customer service folks with a history of handling angry customers automatically get those emails, and so on. Anyway, let me know what you think by leaving a comment below. As a reminder, this is all being done via a so-called “static” site. Pretty damn impressive, right?

Blog

Adding Emotional Tone Analysis to Your Contact Form

Raymond Camden

Join Our Newsletter. No Spam, Only the good stuff.

Related