NewsBalance
Our entry for Facebook’s London regional hackathon 2018. NewsBalance uses natural language processing to find articles of different bias on top news stories, to allow the reader to get a balanced understanding of the story.
The system first searches for the input string using in an api called web-hose. It then starts splitting each input document into it's component words and then running these words through word2vec. We then sum all of these vectors to allow us to create a document vector. We then performed k-means on the set of all documents that showed up in the search, and used the computed centres to come up with our categories. We then stick the input-output mapping in a redis cache to allow us to operate at speed for subsequent queries. A react.js front-end allows the user to search for topics and visually compare them.