Recommendation Systems



Recommendation Systems are models that generate recommendations of items that could be of interest to each user in a personalized way.
Some famous examples of recommendation systems are Amazon (customers who bought this) and Netflix (movie recommendations). Every day there are more services incorporating recommendation functionality for their users. Its rapid expansion is due in part to its wide generality: Getting recommendations is of interest in any service that is available in a user community and a community of items too broad to be reviewed one by one.
In this kind of business model, the main challenge is the automatic management of information for income generation. One of the key systems in the generation of income is recommendation.



Collaborative Filtering:
Recommended systems based on the actions of the user.

User actions give us information about their own tastes and themselves and item information. The underlying idea is that this information can be analysed to capture the following two points:

  • For each user, what their tastes are.
  • For each item, which are the satisfactory tastes.

This idea seems so simple, but requires a complex model with a high calculation cost to be adjusted according to the technical data from the field of Machine Learning.
Its main virtue is that it allows recommending not “understanding” the underlying items. There is no need to have explicit information about them, only the actions of users.





Content-Based Filtering:

Recommended systems based on the actual content of the item.

The information that the item itself has can be processed to infer which items are similar to each other. From these similarities, we can find items where it looks like the user already has a satisfactory relationship, or look to a whole set or subset of items where the user knows it will like.
In turn, there are also hybrid models that combine both.



AIA applies algorithms to generate recommendation systems which:

  • Detect communities of users with similar interests or behaviour.
  • Get a scoring of the importance that each user gives to available products.
  • Allows content recommendation to users depending on the preferences of the user or users associated with it.
  • Allows related content recommendation to those who have expressed interest, allowing the discovery of new points of interest.
  • Allows the evolution study of social preferences and trends.
  • Allows interpretation of social relations translating this into content of interest.