Building a Discovery Engine with Machine Learning

FindLectures.com is a discovery engine for tech talks, historic speeches, and academic lectures. The site rates audio and video content for quality, showing different recommended talks each day on a variety of topics. FindLectures.com crawls conference sites to get talk metadata, such as speaker names and bios, descriptions, and the date a video was recorded. Often these attributes are sparsely populated, or available across multiple websites. Additional attributes are inferred from audio and video content, but require more sophisticated data extraction to be useful in a full-text search engine. This talk will discuss interesting lessons learned from crawling historical videos and demonstrate information extraction with machine learning.