How to use a PBX machine?

What are the ways to use machine learning in a movie recommender system?

• I am new to recommendation systems and have to implement a Movie recommender system. I have previously worked with Classifiers (Naive Bayes, SVM etc.) and understand how I can use Machine learning although I am not able to understand that how would i use Machine learning in a recommender system. I find all the processing happening real time. I am planning to use a hybrid of Collaborative and Content based filtering but can't relate it to Machine learning.

• Answer:

  While you can frame a recommender as a classifier problem -- by building a classifier for each movie or perhaps group of movies -- it is not usually feasible or effective to apply these techniques. Naive Bayes / SVM / random forests etc. don't come into play. You can make recommendations offline or online. It's not necessarily true that you do it all online, though you can try. There is usually some "model building phase" like in any learning algorithm. For collaborative filtering it may be precomputing similarities or clustering, projecting into lower dimensional space, making a latent factor model. CF relies on interactions between users and items rather than the properties of users and items. You can either choose to view these quite separately, build separate models that make separate predictions, and then combine them in an ensemble. Or, sometimes I've seen the user/item attributes used to define a similarity metric and then apply those in a neighborhood-based algorithm that is itself built on the interaction data. That's a way to combine both. I suppose you could even do the reverse! Or you can try to mash them together. For example you could imagine that "user is female" is like "user interacts strongly with an item called 'female'" and then treat attributes like other user-item interactions in a general model. It is crude but actually works well.