HTML + JS + Hazlo.ai: Make your own review scorer

Wouldn’t it be pretty cool if you could make your own review rating algorithm, in simple HTML & JS without worrying about the tedious ML backend infrastructure? That’s exactly what we’re gonna do in this tutorial — build an ML algo which gives ratings to reviews (1–5) in a jiffy.

Dataset

We’ll be using the Etsy review dataset openly available at Kaggle: https://www.kaggle.com/andrewmvd/trip-advisor-hotel-reviews. Credits to Kaggle: Abdulelah for distributing the dataset

This contains the text of a review and it’s score — what rating the buyer gave the seller. We’ll try and predict the score given the review.

Step 1: Upload the dataset to your Hazlo account

Log in to your Hazlo account, go to the datasets page and deploy the .csv file from the link.

Just upload the file, give some context and Hazlo automatically extrapolates, augments and cleans that dataset.

Step 2: Create your project

Once you’ve uploaded your dataset, go to your projects page and launch a project with “ratings” as your chosen column.

Give your project a cool name and choose “ratings” as the prediction column.

After pressing deploy, your model should be ready in a couple of minutes. You can see your models training in real-time and Hazlo takes care of the tedious stuff like synthetic samplings and vectorizations — so we can focus on building our site.

The metrics for all reviews are looking pretty good.

Step 3: Connect your API

You’ll see the sample cURL request for your model in your “APIs & NLP” page; you can convert the request to your language of choice (JS in this case) with https://curl.trillworks.com/ — it’s a nifty tool that quickly converts a cURL request to valid requests different languages.

Your cURL request will look something like this.

& the subsequent JS fetch request should look something like this. The no-distribution parameter enables us to only get the most likely outcome instead of the probability distribution.

Step 4: Build out a front-end

This step’s pretty straightforward, we’ll build out a quick front-end using CSS ( I’ll be using Bootstrap to add that bit of pizazz) & HTML. It’ll need to have a text area input which users can write the review in and a predict button which executes the fetch query.

Here’s what my frontend code looks like — nothing to fancy but gets the job done.

Step 5: Deploy & share with the world

That’s it you’ve built out your review classifier — you can now deploy it and share it with the world.

Here’s my pen, you can put in your project details (project code & API Key) and start using it.

An aside: since this is just a demo, we’re using API Keys on the client side; however, it’s highly recommended that you only use your API key on the server-side of things during production.

“Talk is cheap. Show me the code.”
― Linus Torvalds