What is Recommendation/Suggestion System?
In the articles written on this subject, the word recommendation or suggestion is used. I prefer the word 'recommendation'.
Recommendation System or Suggestion System (in some places 'platform' or 'engine' is used instead of 'system') an information filtering aimed at predicting the amount of 'rating' or 'preference' a user would give to an item system.¹
You can find the wiki definition of the recommendation system above. But since I think this subject is open to interpretation, I am in favor of explaining it with examples.
Recommendation systems are a system that feeds themselves through data and/or user behavior to inform both users and product owners. Today, there are hardly any web pages that do not have this feature. In fact, a substantial part of the income of large-scale companies comes from here (I will talk about this in more detail later). Companies that integrate the recommendation system into their systems aim to keep their users in their ecosystem and continue to interact. For this purpose, companies need to ensure that users have a pleasant and useful time on the platform. Sometimes this can be an alternative to the product purchased or products that are likely to be used with the product², a movie, music, or an advertisement as in our platform.
Although they are generally used to advise users on web platforms, they are also used for product owners. It will be a classic example, but there is a study* called correlation between beer sales and diaper sales. Although it is partly legendary and mostly told about data science, it is possible to realize the same event with the information obtained in the output of the recommendation systems.
*According to this study conducted in the USA, a high percentage of fathers who go to the market to buy diapers also buy beer to reward themselves. With a recommendation such as "who buys diapers, also buys beer", market owners placed a beer cabinet next to the diaper aisle and watched the sales increase.
How does it happen on web platforms?
Online platforms take care to keep users on the site as much as possible. Directing users to the right products is one of the most fundamental building blocks on this path. In its simplest form, someone who buys brushes buys paint, someone who listens to Heavy Metal listens to Metallica, someone who reads Frank Herbert reads J. K. Rowling, and for our world, recommend a different vehicle with the same segment, same price range, and similar features as the vehicle model in our user's mind. It will not be considered absurd by the user to think about the possibility and go in a direction in this direction. On the contrary, the user will be pleased with it, he will easily reach the product he is looking for, and maybe he will discover something new. If we examine the leading companies in the sector, we will clearly see the importance of recommendation systems.
Amazon is a company that has invested heavily in recommendation systems and actively uses many types of recommendation systems within its own body. This diversity, when examined in detail, seems to contribute to the success of Amazon. The company reported that it increased sales by 29% to $12.83 billion in the second fiscal quarter of 2015, up from $9.9 billion in the same period of 2014³. They also get 35% of the total income from their recommendation systems⁴. Amazon currently uses item-to-item collaborative filtering that scales to very large datasets and generates high-quality recommendations in real-time. We will examine this type of filtering in more detail in our next article.
When we examine the Netflix platform, we see that it has built its ecosystem on increasing the number of users by personalizing the experience it offers to increase the profit rate. Although it has millions of subscribers and thousands of visual resources, it aims to provide a personalized experience. It produces thumbnails according to the audience's delight and prepares personalized posters instead of boring posters. Netflix is obsessed with presenting you with content similar to the ones you watch and giving feedback on, based on your mood. It even awarded a $1 million prize to a team of developers in 2009 for an algorithm that improves the accuracy of the company's recommendation engine by 10 percent. If we look at the earnings they expect from the recommendation system in 2016 to understand its value for the company, we can observe that the expectation is 1 billion dollars⁵.
The “Weekly Discovery” product released by Spotify in 2015 is a product designed with a recommendation system is and ML algorithms. Spotify uses collaborative filtering when creating weekly discoveries. “Is the heart symbol also pressed when creating the recommendation list? Which artists does the user follow? What songs are in your user's personal playlist? Which songs does the user dislike?” It uses songs that the user has never heard before, according to the answers to questions such as: The more often you do the above-mentioned actions on the platform, the more the weekly discovery generated will reflect your tastes.
sahibinden.com is an online classified and shopping platform with a monthly visitor count of approximately 60m where businesses buy and sell real estate, cars, and a wide variety of goods and services.
sahibinden.com published the ad recommendation product in 2017. Since 2017, it has included advertisements and recommendations on the home page on mobile and mobile web platforms, and on the search results page on the web platform. sahibinden.com uses collaborative filtering while generating the ad recommendation data. It provides personalized advertisement recommendation output by comparing the advertisements visited, favorited, and messaged by the users in the system. It has an average of 120 million views and 5 million clicks per month.
When we analyze the above examples well, we can observe how great the return of the recommendation system and its importance in the light of this return. In the next step, we will examine the types of recommendation systems with you. Stay tuned😉 🤜🏼
- https://towardsdatascience.com/deep-dive-into-netflixs-recommender-system-341806ae3b48 , https://www.businessinsider.com/netflix-recommendation-engine-worth-1-billion-per-year-2016-6