Collaborative-Based Recommendation Engine Explanation

Collaborative-based recommendation engines are the most popular non-hybrid type of recommendation engine. One reason is because of its ability to capture latent features.

The idea is that similar people pick similar items, regardless of how well we understand why. For example, if someone is from the South in the United States and likes "chicken-fried steak" then they may also like "biscuits with gravy." The two foods are local cuisines.

We could go through an extensive amount of effort to collect local cuisines and extend a huge amount of time doing so, but we would still be less accurate than a collaborative-based recommendation engine that just looks at what people do, not what they say they do.

To illustrate the point, take a family where the mother is from California and the father is from the South. Their family will likely eat foods that are popular in both cultures, but not be pure to either. This would confuse our "local cuisine" ideas. Instead, if we have enough data on what people like to eat then we will find other groups of people that exhibit the preference of both Californian and Southern food and be able to recommend foods from that group.

The biggest problem with collaborative-based recommendation engines is that they require lots of data. They need data of what other people like and what the user likes. Even if we have data on what other people like, how do we get data on new users? This is called the cold-start problem.

DISCLAIMER: This is an example recommendation engine for the purposes of learning and teaching. There are many other features that could be added.