The climax of today was getting to meet the client, engineering news publisher Engineering.com. But while we waited until that meeting, my team mate and I re-watched some machine learning lectures from a Coursera MOOC we took. The lectures describe techniques for analyzing data and making recommendations or sorting the data.
For example, TF-IDF (term frequency-inverse document frequency) helps you recommend similar things (like news articles) based on how relevant they’re calculated to be in the context of every item in a system, and clustering helps you optimize things by pre-sorting the items based on their properties so that the recommender system won’t have to search everything to find the most appropriate recommendation.
After the meeting with the client, we learned about how they want to increase the amount of article reading their visitors do, and target the users more effectively with mailing list emails to help them drive revenue. It sounds like my team mate and I had the right idea as we prepared. We’re going to have to log as much data as we can to uncover new things we can learn about Engineering.com’s users, and understand their behavior. Then, we can help them refine their current article recommender system (which is built into the web app framework they use for their site), or perhaps help them build a new, more powerful recommender system able to cooperate with their current system. Sounds fun!