Current Research

New York Times Article Recommendations

Using user supplied preferences of articles (like or dislike) I built a model using a basic classifier (stochastic gradient descent) to compare two articles (that have different ratings) and predict which one the user would prefer more (using a tf-idf bag of words approach).  It calculates the differences in the tf-idf vectors for pairs of articles and learns the n-grams that are correlated with a user picking one article over another (the methodology was adopted from Fabian Pedregosa's blog, and the the concept of pairwise transforms comes from Herbrich, Graepel, and Obermayer (1999)).  The algorithm uses the title, news desk, snippet, and the full text of the article to rate and recommend articles.  

Anyways, here are the recently recommended articles based on a bunch of articles I "liked" or "disliked."

Please see Most Recent, or see the frame below for the metadata of recommended articles.