Have you ever seen the comment “Congratulations! You have been chosen by the algorithm.” or “An unknown algorithm led me here.” among the YouTube comments? What about the experience of seeing products recommended after a few searches? These are the results of the “Recommendation Algorithm.” We get help or fun through the recommendation algorithm. Not only YouTube but also various fields such as shopping and Information Technology (IT) use it. Typically, Netflix, the biggest online streaming service, sends an email to users when new content that suits their tastes has been updated. This is an example of a recommendation algorithm that hits the taste of various users. Then, how does the recommendation algorithm work and how is it being used in the field?
The recommendation algorithm is a recommending system based on data
An algorithm is a series of mathematical steps, usually in a computer program, which will give you the answer to a particular kind of problem or question. Algorithms are used in various fields such as IT services, manufacturing, and finance as they allow people to process and reflect a large amount of data in real-time. Without making users to make choices after examining tons of contents, service providers can improve users’ satisfaction. The algorithm that is most frequently encountered by general users is the recommendation algorithm that recommends content based on data.
There are two ways in which the recommendation algorithm works: Contents Based Filtering (CBF) and Collaborative Filtering (CF). CBF recommends items based on a comparison between the content of the items and a user profile. For example, it recommends recently released classical record to someone who enjoys listening to classical music. CBF is rarely used nowadays. CF is a family of algorithms which include multiple ways to find similar users or items and calculate a rating based on ratings of similar users. There are two types of CF, which are User-Based and Item-Based. User-Based Collaborative Filtering (UB-CF) is a technique used to predict the items that a user might like, based on ratings given to that item by the other users who have similar tastes with the target user. For example, when user A and user B have similar tastes and if A bought salad, pizza, and coke, UB-CG recommends coke to user B when he/she buys salad and pizza. Item-Based Collaborative Filtering (IB-CF) works based on the similarity between items calculated using people’s ratings of those items. For example, if bread and butter are often purchased together, UB-CF recommends butter when a user buys bread. CF has the limitation that it does not work well with new users as it does not have any data from them. Also, items of low interest may not be recommended due to the lack of data.
How are the recommendation algorithms being used?
Park Chae-rin, a sophomore of Department of Media Communication of University of Suwon, said, “I notice that recommendation algorithm is working while I am using YouTube. I listen to music a lot on YouTube, and it is great that it recommends playlists which usually suit my taste. I often read some funny comments such as ‘I am thankful to the algorithm that led me here.’ on YouTube and they are quite relatable.”
According to the “YouTube Recommendation Algorithm and Journalism report” from the Korea Press Foundation, YouTube does not disclose how the recommendation algorithm works. Still, we can roughly give an idea of the recommendation algorithm’s components and some data used by it from Google’s research papers and externally conducted experiments. To make a list of recommended videos, millions of video data and several algorithms are used. For example, the list of recommended videos add all the videos, which are currently played, with similar themes, and which have been played together a lot, to the list. This course uses the user’s demographic information such as where he/she lives in, as well as the user’s viewing history, such as how long a user has watched a video before and what keywords he/she searched for. After that, the algorithm that determines the recommendation ranking starts working. It predicts and scores the user’s response (involvement, satisfaction) to the recommended videos list. The “involvement” includes the number of clicks and playing time after the click and the “satisfaction” includes likes. (The video playing time is more important to exclude provocative titles and thumbnails aiming at clicks.)
Netflix is making a comparatively transparent disclosure of its recommendation algorithm. The basic user information that is collected is likes, selected and unselected videos, viewing time zone, video viewing devices and so on. In detail, the spot where you rewind or stop watching the video, and whether you watch it until the end are also figured out. Also, Netflix makes “tagger” watch contents in person and tags videos based on the guidelines. For example, when it comes to “Stranger Things,” its genre was classified as SF, high-teen, thriller and horror, and was tagged as ominous energy, scary stories, and interesting.
The recommendation algorithm significantly affects our lives
The recommendation algorithm has the definite merits of satisfying the needs of people and giving unexpected joy. However, as most service providers use it to provide personalized information, the users will only be exposed to the information that is filtered. This phenomenon is called the “filter bubble.” In other words, the recommendation algorithm has the advantage of obtaining personalized information that reflects users’ tastes, but Internet information providers can strengthen stereotypes and prejudices by focusing on what individuals like.
Park Chae-rin said, “The recommendation algorithm is working the most when some commercials are popping up. Recently, I was interested in vintage clothing and searched some on Instagram. Since then, some commercial accounts related to vintage clothing kept appearing. Like this, I have often felt convenient as I do not have to look for information thanks to the algorithm working instead of me.” Also, she added that it is convenient when we take advantage of the principle of algorithm. “Once, I wanted to look for some winter clothes and it bothered me a lot to look them up one by one. Therefore, I intentionally pressed likes of some shopping mall accounts and clothing commercials kept popping up.”
Meanwhile, some people are worrying about whether too much of their personalized information is exposed. Park Chae-rin said, “It is not that unacceptable for me as our information is already used widely to the extent that there is a joke ‘Our information has already been spread throughout the world.’ Also, I became more convenient by using algorithms. However, I am also worried that my private information could be abused and hurt me. Still, I enjoy the recommendation algorithm because it makes my life more convenient.”
The recommendation algorithm is being utilized in various online systems, including video platforms, shopping sites, Social Networking Service (SNS), and music services. How well the algorithm recommends suitable information among tons of data to users is an important factor that determines the success of online systems. Therefore, lots of online systems are trying to improve existing recommendation algorithms and provide better services. It is expected that much more suitable recommendation services will be provided to users in the future.
Lee Su-yeon email@example.com
<저작권자 © 동국포스트, 무단 전재 및 재배포 금지>