What’s Poppin

landing page to input user query
globe with lights that show where news are metioned
articles from specific locations with images
when no articles are found

Inspiration

What is going on in the world around us? What are people talking about? We wanted an interactive way to see every single thing that was happening on Earth right now. We wanted a way to see how people in different countries experience the news—how stories are written for local audiences and what is truly happening from their perspective. That’s why we built a platform that lets you navigate a 3D globe and discover news articles from around the world, covering the topics that matter most to each region.

What it does

It’s an interactive 3D globe that lets you explore popular news articles, shown as pinpoints in the regions where they originate. On the landing page, you can search for a specific topic or filter by categories like politics, technology, or sports to discover stories from around the world.

Once you enter the interactive map, you can zoom in and out to refine your search, revealing more trending and locally relevant news articles within each region.

How we built it

We used GDELT to pull free public news data, and filtered and processed it. This involved over 1000 articles every 15 minutes, and we ran this on a server continuously to pull data. Because the news articles were from around the world, they had to be translated into English among other processing methods. The processed data was then saved into a supabase to store all values like longitude, latitude, article title and then used all-MiniLM-L6-v2, an open-sourced hugging face model to embed this information into a 384-dimensional space and also store that into supabase. Then we created a search function and suggestions as well as rendered a globe using Three.js in the frontend. To match desired results with relevant news, we used all-MiniLM-L6-v2 again to embed the search query and used K-nearest neighbour search algorithm to determine the most relevant news based on proximity of the embedded search query to the embedded news articles. We then took those news articles, populated them onto the globe based on its latitude and longitude, and rendered the articles and images on the side.

Challenges we ran into

It was difficult doing the eda on the data and ensuring that it returned relevant information. In addition, rendering the frontend was a constant struggle.