Inspiration
I mainly just wanted to play around with the Cohere API and explore text embeddings more.
I did some work analyzing votes from UNGA resolutions over the summer and found it hard to read the long and verbose contents, and I also found the search system confusing at times, especially given the non-descriptive titles.
What it does
Two main functionalities:
- Summarizes UNGA resolutions in the tone of a second grader
- Calculates embeddings for each resolution so it can find closest ones to query during search
How we built it
Mostly following Cohere API documentation and lots of StackOverflow posts!
Challenges we ran into
Never enough time to do all the cool things there are to do!
Accomplishments that we're proud of
This notebook works well as a proof of concept for some future work :)
What's next for UNGA Resolutions: Search & Summarize
- Look into clustering analysis more on actual labels and proposing countries
- Do one step further on search and actually include relevant information in prompt to answer question directly
- Run this for more than just 96 resolutions, there are so many out there!
- Improve the text summarization prompt and/or fine-tune the parameters
Log in or sign up for Devpost to join the conversation.