Ji-hye’s Happy Dilemma
There is a newly opened library in a small town. Ji-hye, the librarian, has one goal: to ensure that everyone who visits the library discovers their own ’life-changing book’ and leaves satisfied.
However, people’s tastes are incredibly diverse. What book should she recommend to truly resonate with them? Ji-hye found herself deep in thought as she flipped through the library’s loan records. It was there that she discovered two very interesting patterns, which are the core principles of Collaborative Filtering that we will discuss today.
First Discovery: Finding a ‘Soulmate’ with Similar Tastes
The Connection Between Min-jun and So-ra
While examining the loan records, Ji-hye was surprised to find that the book lists of two patrons, Min-jun and So-ra, were remarkably similar. Both loved fantasy novels and enjoyed historical mysteries.
One day, as Min-jun borrowed a newly arrived fantasy novel, he exclaimed, “This book is amazing!” Hearing this, Ji-hye immediately thought of So-ra.
‘Ah! If Min-jun liked it this much, So-ra will surely love it too!’
When So-ra next visited the library, Ji-hye confidently recommended the book, and So-ra became an enthusiastic fan of it.
This is ‘User-based Collaborative Filtering’
What Ji-hye just did is the basic principle of User-based Collaborative Filtering. In simple terms, it recommends ’things that other people with similar tastes to yours have liked.’
- Netflix Example: Let’s say I enjoyed the thriller movie Searching and the sci-fi film Interstellar. Netflix identifies many other users who rated Searching and Interstellar highly, just like I did. If those users commonly enjoyed another movie, say Find Me, Netflix would recommend it to me. Even if I had never heard of that movie, it has a high success probability since it has already been validated by ‘people with similar tastes.’
- YouTube Example: Suppose I frequently watch ‘cat’ videos and subscribe to ‘game streaming’ channels. YouTube finds other user groups that enjoy both ‘cats’ and ‘games,’ which may seem like different categories. If that group starts watching ‘cooking’ videos, one day, a ‘Baek Jong-won recipe’ video might suddenly appear on my YouTube homepage. This is the power of user-based filtering, which recommends new interests based on not only my direct actions but also the behaviors of similar users.
Second Discovery: Finding ‘Best Friend Books’ that Attract Each Other
The Secret of The Little Prince and The Alchemist
This time, Ji-hye focused on the books themselves. She discovered that people who borrowed The Little Prince often borrowed The Alchemist shortly after. Conversely, those who were deeply moved by The Alchemist tended to seek out The Little Prince.
It was as if the two books were saying, “We are perfect friends!”
After that, whenever someone borrowed The Little Prince, Ji-hye naturally asked, “Have you read Paulo Coelho’s The Alchemist? If you liked this book, I’m sure you’ll enjoy it.” Surprisingly, this recommendation was mostly successful.
This is ‘Item-based Collaborative Filtering’
Ji-hye’s second discovery illustrates the principle of Item-based Collaborative Filtering. This time, the focus is not on people but on the relationships between content, or items. It recommends ‘other things similar to what you liked.’
- Netflix Example: Netflix’s ‘Similar Content’ list is a prime example. If I binge-watched Stranger Things, Netflix analyzes what other content many viewers who watched Stranger Things subsequently viewed. As a result, it finds that series like The Umbrella Academy or Black Mirror are often watched together. Therefore, it recommends these series to me, not because they share the same director or actors, but purely based on the strong correlation found in ‘consumption pattern’ data.
- YouTube Example: On YouTube, when you finish watching a specific video, the ‘Next Video’ list that appears on the right or the recommended videos below the video actively utilizes item-based filtering. For instance, if I watched a ‘IU live clip,’ the system analyzes what other videos were immediately watched by users who also viewed this video, such as ‘Taeyeon’s live clip’ or ‘other singers covering IU’s songs,’ and recommends them as the next video for me to watch. In this way, the item ‘IU live’ and the item ‘Taeyeon live’ have become best friends in users’ viewing records.
The Invisible Librarians Among Us
Smart librarians like Ji-hye are everywhere around us. Netflix and YouTube skillfully mix these two collaborative filtering methods. Sometimes they find a ‘soulmate’ similar to me and borrow their choices (user-based), and other times they introduce me to ‘best friends’ of the content I’ve watched (item-based).
Now, when you hear the term ‘collaborative filtering,’ how about thinking of the kind librarian ‘Ji-hye’ who helps find someone with similar tastes or a best friend for the book you love? Although the name of the technology may sound a bit cold, its essence originates from a warm intention to connect people’s joy and satisfaction.