Jasmin Rubinovitz


An interface for smashing filter bubbles, Panorama is built to allow open, transparent, and collaborative exploration of news from all across the political map. It presents different perspectives and encourages serendipity in news exploration, versus getting all of our news from one single source. Panorama is a human-in-the-loop interface. The computer processes more than 10,000 news stories each day, both broadcast and written, and it uses machine learning algorithms to decide what topics each story is talking about and if the stories are positive, subjective, or trending. The machine learning process pours over massive datasets and learns to generalize in smart ways, but not in the same smart ways that humans generalize. As a result, it can be brilliant and also get very confused. With Panorama, some of the training data was a large open set of movie reviews, and while this is a great dataset to start with, it is not mapped so well to news stories. As humans interact with Panorama, they are encouraged to give better labels to stories; those labels are fed back into the algorithm to make it better.

Having a lot of information about each news story and all stories together allows us to create an open-box news aggregator. With most aggregators we use today (like the Facebook News feed), the user has no idea what are the algorithms and filters that decide what s/he will see. Panorama is open: the user can decide to view everything, or filter only to specific things that he s/he is interested in, by playing with the sliders and seeing in real time how the news feed changes accordingly. For example, you could easily get all stories about animals, from the right side of the political map, that are also positive and objective. Panorama also exposes interesting patterns, such as the topics that different news sources focus on every day, and what sources had many objective versus subjective stories.