Have you heard of recommender systems?

Have you ever had the strange experience where you're talking to a friend about seafood pizza and then every ad you see on the internet after that conversation is for seafood pizza, or in some way related to pizzas? This happens because of recommender systems.

Recommender systems use supervised and unsupervised learning approaches to learn about us. Since we're so multifaceted, recommending things to us is a difficult task that may yield a lot of unexpected outcomes.

Therefore, recommender engines use artificial intelligence (AI) to understand our minds and provide us with valuable recommendations. The same technology is used to suggest YouTube videos or shows on Netflix.

AI that can make suggestions has the potential to alter the internet drastically. However, to comprehend the benefits and downsides of these algorithms fully, we must first learn where they acquire their data and how they operate.

The three main techniques for recommender systems are content-based recommendations, social recommendations, and customised recommendations.

Content-Based Recommendation

The substance of the content is the focus of content-based recommendations, not the audience. For example, the algorithm may choose to promote more recent videos or films created by someone on a list of excellent creators. Information such as items or services will be returned based on your preferences or viewpoints. The more information the user provides, the better the accuracy of this recommendation.

Personal metadata and individual transactional data may be absent at the start of some services due to privacy and legal concerns. For recommender systems that use this method, these concerns are frequently referred to as "cold start" issues. A cold start happens when a recommender system is unable to make conclusions for a query owing to a lack of data.

Social Recommendation

Social recommendations pay attention to the audience. It utilises social metrics such as likes, views, and view time to determine what people are watching and engaging with, and should be recommended. This type of recommender uses collaborative filtering, by utilising their judgement and behaviours to make product recommendations to you or determine how one product could complement another. A common example is "next buy" recommendations while online shopping.

The problem here is that in the absence of previously acquired data, they are particularly vulnerable. It gets more difficult to participate in any single-person action when there is no relevant knowledge on others.

Personal Recommendation

People have various tastes, which AI systems can take into account when making tailored suggestions. The difficulty with customised suggestions is that it may be difficult to come across new, fascinating content, therefore recommender systems often use collaborative filtering, which combines all three of these recommendations to obtain the best of both worlds.

This notion that we all view slightly different versions of the internet and that data is continuously collected about us might be alarming but knowing how recommender systems function can help us live more informed lives while coexisting with AI.

References

1. Oracle & DataScience.com. Introduction to Recommendation Engine.

2. Shetty, B. (2021, June 2). An In-Depth Guide to How Recommender Systems Work.

All Nashrah cares about is smashing the patriarchy. Help her at nashrah.haque01@gmail.com