Problem Formulation

  |   Source
Problem Formulation

  • Recommender System is one of the most important application in Machine Learning.
  • When Ng gather information from various company in Sillicon Valley, often he asked what is the most important problems in implementing Machine Learning, he get answered on how to improve the performance of Recommender System
  • Recommender System is usually used by Amazon, Google, or any kind of online store or others that trying to recommend based on the user purchased or rating to them.
  • Nowadays, Recommender System has a little intention among Machine Learning practioners, even though it raises many problem for the companies in Machine Learning
  • That's the main reason, the other is as we approach the end of the class, by now we should know that choosing the right features in Machine Learning is a huge impact for the learning algorithms. Some features tend to make it bias, some distinguish anomaly from normal. Either way, choosing the features manually may not a good idea, in addition if we have to choose from thousand of features.
  • This lesson would teach us how is the problem we tend to have in Recommender System



  • The slide above is one of example in Recommender System, that is how's the user rating previous movie they watched, and try to recommend the movies they never seen before.
  • nu = number of user, nm = number of movies. for '?' rating, means that either they don't rate it, or they haven't watch it yet. Safe to assume they take the later option
  • r only equal 1 if user has rated the movie
  • Recommender Systems is trying to guess the movies rate they haven't watched it before
  • And based on all movies that they rate, our job in Recommender System is try to come up with a guess of what rating they should give for movies they haven't watched.
  • And based on all the rating the algorithm guessed, pick movies with high rating and recommend the users the movies that have high rating prediction.
  • SUMMARY
  • Next video, how to handle the problem of recommender System