Product update: Thematic’s sentiment analysis more accurate with a 20% reduction of error

Nigel Warren
Nigel Warren

This is the second product update we’ve shared on our blog. So far we’re finding this medium a great way for our customers and enthusiasts to keep up to date with what’s new within the platform.

Our team is always working hard to improve Thematic, and today we have some exciting news to share from our R&D department, as well as product.

Let’s take a look!

Thematic is launching custom sentiment models!


Our R&D team is always looking for ways to make Thematic more accurate.

After resolving some technical challenges, our fantastic R&D team has made it possible for us to support custom sentiment models!

In the past Thematic worked with one general feedback sentiment model... We tuned it on millions of pieces of feedback and even proved that it’s more accurate than Google’s model.

However there were two types of errors this generic model made:

  1. Occasionally, it didn’t know the specific meaning of a word within a particular context. For example, “lying" is usually a negative word. But, when the user shares feedback on "lying comfortably in a bed", this word is neutral.

  2. When people rave about how "Product X is amazing", it's positive sentiment if you are the maker of this product. If Product X is made by your competitor, the sentiment of the comment is likely negative.

So how is a custom sentiment model different? A custom sentiment model tailors the analysis to capture these nuances. It can lift the accuracy by up to 15-20% depending on the dataset.

Improving the accuracy of your sentiment analysis means the reliability of your insights also improve.

Tailoring your data on a custom sentiment model is a paid service we now offer to our enterprise customers.

We’ve streamlined our setup flow

We’ve made it even easier to add new datasets to Thematic. There's a fresh new UI, and we automatically detect key data types so that you can get your new analysis started quickly.

Step 1. Name your data, import from a file or one of our many integrations, and continue. Simple!

The import data stage of the Thematic setup flow
The import data stage of the Thematic setup flow

Step 2. Review your data! We now allow you to specify whether your first row contains header labels so that they aren’t included in your analysis. We also automatically detect comments, scores, and dates so all you have to do is review and click continue.

The review data stage of the Thematic setup flow
The review data stage of the Thematic setup flow

Step 3. Sit back and relax as Thematic does the heavy lifting analyzing your dataset.

You can now review your untagged comments

A new addition to our themes editor is the ability to review your untagged comments.

In a dataset analysis there is typically a small percentage of comments that we deliberately leave untagged. These include:

  • Comments containing gibberish, like random strings of letters and nonsense words.
  • A very small percentage of comments (often only 1 comment) that don’t match any existing theme.

With this new functionality, you can now look for feedback that isn’t already captured in your themes.

You can find the button to review your untagged comments at the bottom of the themes editor.

Once clicked, you will see your untagged comments. You'll also see the approximate coverage of the comments tagged during the analysis, handy!

The review untagged comments section in Thematic
The review untagged comments section in Thematic

If you find untagged comments that you want to tag, simply add a relevant phrase from the untagged comment to an existing theme. Or, create a new theme altogether.

Here’s a quick example: You have 2 separate untagged comments, one reads: ‘No native analysis of Arabic’. Another reads: ‘No native analysis of Portuguese’. You could create a new theme ‘Native analysis in other languages’, and tag these two comments within this new theme.

We hope this new feature helps you capture straggling comments and strengthens the quality of your analysis.

Sometimes your comments in Thematic come packed with extra data, such as an ID number for a customer in your CRM, or a support ticket in Zendesk.

To save you the hassle of looking up this ID in your own systems, we can now make these clickable so that you can quickly move from Thematic to any other internal systems you have.

Feedback in Thematic has hyper-linkable meta-data
The TICKED ID is hyperlinked

We hope this saves you valuable time!

Contact our friendly product specialists to learn more about these features!

Using ThematicProduct Updates

Nigel Warren

Nigel is the director of UX & Product at Thematic. Working tirelessly to bring a delightful experience to the users of Thematic.


Table of Contents