The Gadfly Project

A simple service that instantly enables developers to automatically generate questions from input text.

Today, the VANDALS announced the release of the Gadfly[1] Web API that powers the Twitter account, @question_exe, which has been tweeting intriguing questions about articles in the New York Times. The API unifies a set of tools that make it easier for developers to generate questions based on input text without the headache of dealing with NLP libraries. The diligent VANDALS team plans to provide support to make integration as seamless as possible.

The problem of question generation from text is an interesting one addressed by Heilman and Smith in 2011. The VANDALS team, building on that work, has created a web service that uses proprietary algorithms to process text from a news article to generate questions about that article.

“Richard Feynman famously said, ‘There is no learning without having to pose a question.’ This is our attempt at posing questions based on the news to inform readers,” said Daniel Griffin, one of the creators of The Gadfly Project.

The team thinks there are several possible applications for such a service. One of which is their conversational interface, QBot. “It is like nothing I have ever used. It’s made reading news much more interesting,” said Steve, one of their beta testers. Simply put, QBot asks you questions about the news. “Sometimes, it’s just like fill in the blanks, you know. I personally like to use it as a trivia app. Time just flies on BART,” said Jenn, another beta tester. The team is looking to launch this in a few months.

For now, you can visit Paste a news article URL in the text box and start asking questions.

“We were interested in the idea of using question generation as a way to improve online discourse around news. Our initial idea was that we would generate questions based on the news article to prevent people who haven’t read it from commenting on it. However, we found out through user research and customer interviews that this problem is larger than we envisioned. The news publishers that we spoke to were interested in the idea but they had reservations about adopting it. We conducted a public brainstorming session during InfoCamp 2016 and realized that question generation itself is quite an interesting problem. We decided to make the process [using algorithms to generate questions from text] easy, fast and scalable. We have had interest from some companies in the education and news sector. We are genuinely excited about exploring the potential uses for our service,” the team said in a joint statement.

[1] A gadfly is a person who interferes the status quo of a society or community by posing novel, potently upsetting questions, usually directed at authorities as per Wikipedia. The term is originally associated with the ancient Greek philosopher Socrates, in his defense when on trial for his life.

Thanks to Vijay, Andy, Daniel, and Anand for their work on The Gadfly Project which you can try out at here. Vijay, Andy, Nikhil (that’s me), Daniel and Anand are the VANDALS mentioned above.

We were at an odd stage in our capstone project (we had done work towards achieving our goal but our success criteria was not well defined). As a team, we decided to work out what it is that we wanted to create as deliverables by the end of our semester. I decided to use the “working backwards” approach which I learnt over my summer internship at Amazon. You work backwards from the customer to identify what you want to build. A product manager, ideally writes out a document that lists out the customer problem, how current solutions fail and how the new one will succeed. I modified this format to talk about what we were creating (since there are no current solutions that do this), potential benefits we would create for our users along with use cases that we could actually present as deliverables. We are currently building not one but two Slack bot users and constantly improving our API. Stay tuned for more updates on that!

Thanks to Alex for reading a version of this and providing feedback.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.