TED 2.0 is a web-based system that automates the process of creating a TED Talk-esque speech.
What it does
Inspired by the TED A.I. XPrize, an upcoming competition focused on the automatic generation of a TED Talk, TED 2.0 seeks to design and build a proof-of-concept that will eventually compete for the XPrize.
Ideally, the TED 2.0 system can accommodate any user, from those preparing for a speech to those who simply want a TED Talk-like speech on a specific topic.
Users simply visit the TED 2.0 website and select a topic. In less than two minutes a speech will be generated. The system also returns news links and current TED talks based on the topic, an in-app dictionary and thesaurus tool, and an avatar that can deliver the speech.
How it works
The system forms a thesis by picking a debate from debate.org and combining the arguments presented by commentators on that debate. It then extracts keywords from the thesis and uses Google’s related search to find relevant content for the other sections of the talk (e.g. “Importance”, “Problem”, “Solution”, and “Impact”). It applies filters to the content like length constraint, keyword match, and avoid duplicates, to find good paragraphs. Transitions are added to the paragraphs to improve the flow of the talk and a relevant quote is added to the end.
- Develop an active supervised learning framework to improve the quality of the talks by learning to distinguish between good and bad TED talks.
- Incorporating personal anecdotes to make the talk more engaging.
Student team: Amanda Anumba, Divir Gupta, Arundhati Jaswal, Yang Zhang
Faculty guidance: Larry Birnbaum and Rich Gordon