Recommendation Systems and how they work
Recommender systems at their core are just automated systems to filter some entities. These entities can be any products, ads, people, movies, or songs. We see this regularly from YouTube to Netflix to Amazon to Pandora. For example, we watch a movie and then later get a recommendation for a different movie based on the power of previous viewing history. Or it could be a product that we purchased and we get recommendations on other products based on previous product viewing or purchase history. The system doesn’t only work in what products are shown, but also in what order the products are being ranked.
Why are recommender systems being built in the first place? Businesses are showing relevant content for multiple reasons. First off, most businesses think they understand their customer, but often times the customer can behave unpredictably. It is important to show users what is relevant to them while also sharing new items they would be interested in. Recommender systems also serve to help us solve the information overload problem and help us narrow down the set of choices. Businesses get the benefit of selling more relevant items to the user. It is also there to help the customer discover new and interesting things, helping them save time.
How does it work? A recommendation engine will either ask you what you want, whether the content is relevant, look at other users with similar behavior, or study your activity. In some cases, machine learning algorithms can discover more about a person than other humans can. For example when we go on Netflix, or any other service that relies on recommendations, the first time they will ask you what your taste preferences are. That way they can avoid a “cold start problem”.
Let’s say we place an order on Amazon for apples. Now we will be shown apples slicers in the recommended products. This apple slicer has over 10,000 reviews and a five-star rating. Number of reviews, user ratings, and relevancy can all play a factor in terms of what is being recommended.
Another example is YouTube’s suggestion system. Videos that have high watch time, engagement, relevant to the topic being searched, and relevant to the search history will be shown to the user. If the user watches a trailer for ‘Avengers Endgame’, they might be shown a trailer for a similar movie like ‘Iron Man’ or bloopers from the filming of the movie or even an interview with some of the actors.
These algorithms can be so smart that they are able to decide what to show us with scary accuracy. It should be noted that each company might have their own algorithm and way of generating recommendations. One company’s method for making suggestions does not apply to all.