Adding Azure Text Analysis to Fiskkit.com Article Insights

Inspiration

Our inspiration is the terrible state of our public discourse, both in terms of how accurate information is that people put online, and also how people treat each other. We want to use analytics and design to build a better public discourse.

What it does

This demo is an extension of Fiskkit.com, and public discussion tool. This demo (rates the sentiment of each sentence in the article / rates the aggregate sentiment of all of the comments on each sentence in an article). It also identifies Named Entities in an article submitted by users, shows the aggregate sentiment of all the sentences the entity is named in, and allows the user to click on the Entity to toggle highlighting the sentences in the article where that entity appears.

How we built it

First we looked through the available APIs and selected Text Analysis as something that would be really helpful to the Fiskkit project. We did an initial design in Figma and refined it in a team check-in. We sent each sentence in the article and created a list of Named Entities of Type Person, Location and Organization. In the first version, we fed the sentiment results to the frontend to display in-line of the article on the Fiskkit Insight Page. Then we averaged the sentiment scores of each sentence containing a Named Entity and showed that at the bottom of the page. Then we got all of the immediate child-comments of each sentence in the article from Fiskkit’s comments, and sent those for sentiment ratings, and displayed those in-line in the article. Lastly, we set the Named Entities list at the bottom of the article to be clickable so they will toggle highlighting of the sentences that contain that entity.

Challenges we ran into

Because we were sending every sentence of an article for sentiment rating, as well as all the aggregate text that contained each named entity, and all the comments in Fiskkit for each sentence… we ran short on API calls and had to switch to just a couple sentences towards the end.

Accomplishments that we're proud of

We scoped our goals pretty well and achieved all of them in one weekend (plus a little more). We also think this is an actually very useful piece of functionality to add to a public discussion for social good!

What we learned

We learned that it is really quick and easy to use such off-the-shelf 3rd Party APIs as long as they are well-documented, and this will inform our thinking going forward.

What's next for Adding Azure Text Analysis to Fiskkit.com Article Insights

We want to test the accuracy of the sentiment ratings of the API on a larger data set, and also iterate the UI design to see what is the best way to present this information to users.