At the Neural Information Processing Systems Conference in Vancouver, Ralf Herbrich and Jurgen van Gael of FUSE Labs showed the latest version of “Project Emporia” — an all new HTML5 UI for starters.
Project Emporia is a recommendation engine for news. Based on the Matchbox technology from Microsoft Research, it uses a Bayesian probabilistic model to learn the preferences of users for recent news stories. Emporia recommends news for you based on news you’ve read earlier — it predicts what you may want to read. When using Emporia you can vote a link up or down and that influences what you continue to see.
“Emporia mines RSS feeds and all links shared on Twitter — discovering around 1,000,000 articles every day. It does automatic classification of articles into categories using another MRT for classification. MR also developed a system for “active learning” to automatically discover links that cann’t reliably be classified. These types of links are automatically sent to Amazon Mechanical Turk for labeling, they’re then “spam filtered” and returned the classification model to be appropriately categorized,” explains Steve Clayton.
More Info: Project Emporia