Proceeding with various methods for news recommendation

Question

I am beginner in ML (i have done only Andrew Ng's ML course) and i have to work on news recommendation.

I went through this paper which mentions different methods used for news recommendation (at 7th page) and most of them are using some sort of probabilistic methods (bayesian networks, latent dirichlet allocation, naive bayes model, probabilistic matrix factorization models). Also, some news recommenders are based on multi-armed-bandit problem (e.g. yahoo's front page). I have zero knowledge about these methods.

I am very confused regarding what should be my next step. Right now, i am planning to go ahead with a very basic hybrid approach (with collaborative filtering and content based filtering). But it seems like i will have to explore these fields very soon, so what will be a good strategy (and resource) to explore these fields?

Answer 1

Recommender Systems are a huge topic of its own right and goes without saying, with a lot of research going on.

This book does a deep-dive into recommender systems and may not be something you want, but it's helpful as a reference. It seems like you were unsure what those terms mean. The Berekely AI Course covers most of these topics and their lectures are available for free.

You might be able to make better choices after going through some of the above materials.

Answer 2

The best way to decide a good strategy will be to look at this in the two following ways:-

• From a scoring approach(Not Recommended): You can test all the aforementioned algorithms with your data sets and check which gives the best score. You can use tools like Sci-Kit Learn to implement all those algorithms and which has the best score as corresponding to your data set and use that algorithm. The disadvantage being that sometimes even of the score is very high the data set might not be coherent with that algorithm with that I mean you might not get the results you want despite having a high score. This is why I don't recommend this.
• From a dataset oriented approach: For this approach, you have to have carefully studied your dataset. Even starting off with content based filtering isn't a bad idea. But depending on your needs and your necessity to improve precision in specific cases will ensure you explore algorithms to solve that particular problem. This way you won't stress yourself with all the unnecessary algorithms. You can start off using Naive Bayes as it is simple to understand and then pick it up from there depending on what you need in your recommendation, what are the features you are using to provide recommendations. All such factors come into play. This way you can explore the required methods you need to make your model more efficient.