Jtdylktuy

I am studying about evaluation of both recommendation systems and machine learning algorithms in recent times, trying to define a scope for my masters research. After some reading time I'm starting to understand several concepts, but one thing was not clear to me:

Do recommendation systems necessarily use machine learning algorithms?

I mean, I know these two can be used combined, but in most of the papers I read about recommender systems evaluation, they do not even mention anything about Machine Learning.

Also, if you can suggest some papers that I can read, I would be very grateful

There's nothing about a recommendation system that absolutely necessitates some kind of machine learning. Indeed, I've seen decision systems in use that were essentially just someone's idea about what the customer's preferences ought to be.

A recommender can be based on anything from a few ad-hoc 'common sense' rules, to a logistic regression someone did on some data a few years ago and whose parameters are hardcoded into the system, to a complicated ensemble of machine-learning algorithms that are regularly and constantly trained on new data.

The use of machine learning for recommender systems is partly driven by necessity, partly by fad (at least from what I have seen). If a simple recommender works well, and accurately predicts what the user wants, there's no need for a machine to learn anything. If there's a huge amount of data, hiding some very deep relationships that humans are unable to pick out, that's where machine learning becomes useful.

The recommendation system is a broad term to describe everything from a poster "in case of fire.. " to ML-based systems that continuously evolve over time.

A simple recommendation system consists of:

• Knowledge database representing some wisdom that will be used to make new recommendations. This database can be created from historical data, a model, or simply invented.




• Recommender engine - some sort of logic that takes inputs, run them with the knowledge database and produce a recommendation. This can be simple instructions e.g. before crossing a street, look left, then right. It can also be a decision tree helping to identify the best course of action, or a trained ML classifier. It can also be a generative neural network that takes user input and generates new stuff. For instance, predicts text as user types or suggests other books based on recent purchases.

An alternative term for a recommendation system is an expert system. The heyday of these systems was the eighties and nineties. I suggest you look for older papers and books.

Nowadays, machine learning is on the hype and often used where a simple decision tree would suffice.