Rebirth 16: From Hijacking Douyin to Dominating the World
057 One step ahead, always ahead
Now, Lao Ding regards his boss as Douyin's product manager and a master of recommendation algorithms, rather than a 19-year-old boy.
He looked up from behind the screen full of code, his dark circles under his eyes contrasting sharply with his excited state of mind.
He rapidly switched screens, brought up key data panels, and spoke at breakneck speed:
"Mr. Xia, you've come at the perfect time! The data has been running for a week, and it has basically verified the correctness of your core idea. The dual-tower model of 'long-term interest + real-time interest' is indeed working, and user stickiness is even better than we expected."
He pointed to a continuously rising curve:
"Look, the core metrics of average video views per user and average user session duration are both steadily increasing."
In particular, the "Drop Tower" topic that we strongly promoted saw very high tag spread efficiency, proving that content-based filtering was successful.
Xia Wei leaned closer to the screen, carefully observed the data, and asked a question:
"When I was browsing, I felt that the recommended content was too biased towards my real-time interests. Once I liked a work, several similar works would be recommended in succession."
Ding Ming stroked his chin, clicked on another visualization, and pondered his answer:
"Our interest recall model is designed as follows."
"Can you make some changes? Otherwise, users will quickly get tired of the look."
Ding Ming remained silent, carefully analyzing the data.
Data doesn't lie, and it speaks for itself.
In the eyes of data analysis experts, every user action is traceable.
Data shows that some users' completion rate and interaction rate dropped rapidly after continuously viewing similar content.
How to solve this?
Ding Ming pondered for a moment, then a flash of inspiration struck him, his spirits lifted, and his voice involuntarily rose by half an octave:
"In this way, we forcibly incorporate the 'diversity breaking' rule into the sorting strategy."
For example, content from the same creator cannot appear more than three times in ten consecutive recommendations.
"Content with the same tag cannot appear more than once in three consecutive recommendations. This forces users to consume more diverse content."
Xia Wei nodded in approval, "This is a good solution. Let's keep it this way and improve it in the next version."
"Okay," Ding Ming pulled out his notebook and wrote down the newly generated "brainstorming" ideas so he wouldn't forget them.
Xia Wei stated the main purpose of his trip:
"In the next version, we will add more user behavior tags and increase the weight of favorites and searches."
The first version of the algorithm was very rough due to the rush to launch, and continuous optimization and refinement are necessary.
Ding Ming understood the action of adding user behavior tags, but not the second half of the sentence. He frowned and asked, "Why?"
Xia Wei explained:
"The act of saving often indicates that users believe the content has long-term value, rather than just providing instant gratification, which is more meaningful than simply liking it."
"A user's proactive search behavior is a strong signal of expressing a clear interest. If a user searches for a certain term, the interest expressed by the search should be quickly incorporated into the recommendation model in the subsequent recommendation stream."
This is Xia Wei's role.
He can use simple and easy-to-understand language to guide Ding Ming in improving, modifying, and enriching the recommendation model, making personalized interest recommendations increasingly accurate.
The algorithm for "True Fighting Sound" was also gradually modified and refined in this way.
When it first came out, it was even cruder than the first version of the algorithm Xia Wei made; it was simply unbearable to look at.
Because it is a character engineer's model of short video users based on assumptions and predictions, it contains many errors and omissions.
Xia Wei, however, had the memory in his mind and directly told Ding Ming the accurate user model, allowing him to implement it step by step from easy to difficult.
In layman's terms, it means:
The first version only gave him 5 key user behavior tags, such as: completion rate, three-second bounce rate, and like-to-view ratio.
In the second version, he was asked to add two more key actions: bookmarking and searching.
This process of evolution and iteration continues in this way.
However, not a single step in the process was wrong; all of it was key data that had been verified in a previous life.
Xia Wei was born to stand on the shoulders of giants, copying standard answers without going through the process of trial and error and correction.
Therefore, the current algorithm is much more accurate than the algorithm of Zhendouyin—A.me.
Xia Wei estimates that the current level of this first version of the algorithm is equivalent to the level of A.me in the summer of 2017.
They were originally planning to release the first version by the end of the year;
When it first launched, compared to "Volcano", "Watermelon", and "Neihan Duanzi", it was a nobody.
The company was also unsure about its development direction and didn't allocate much traffic or human resources to A.me.
This resulted in extremely slow development and a small user base in the first six months, leaving no user data for learning.
Only after a certain number of users have been accumulated and the algorithm has been iterated and optimized will it gradually show its power.
The set of tools that Xia Wei brought out in advance was more than a year earlier than in his previous life, and several versions ahead of time.
This can be described as a dimensional reduction attack within a dimensional reduction attack.
Don't think of Douyin as something magical; frankly, it's worthless.
Its strength lies in its ability to continuously evolve through deep learning from massive amounts of user behavior data.
In the competition among short video apps in the past, A.me had a first-mover advantage, being the first to use a recommendation algorithm and continuously iterating on it.
The more users you have, the more accurate your recommendations will be, and the greater your lead will be.
In this life, Xia Wei will replicate this advantage.
One step ahead, one step ahead.
This makes it impossible for those who come after to catch up.
The core issue is that Xia Wei possesses a triple barrier: "time difference + cognitive difference + implementation difference";
These three advantages are interconnected, forming a moat that is difficult for newcomers to overcome.
In terms of time difference: it allows the algorithm version, which is due a year later, to be released ahead of schedule.
Cognitive gap: It is a reduction in the dimension of cognition.
He skipped all the trial and error stages and directly broke down the algorithm logic and user behavior model that had been verified by hundreds of millions of users in his previous life into steps that Ding Ming could implement.
It's like teaching Ding Ming how to answer questions using the standard answers.
In terms of implementation: the first-mover advantage lies in the algorithm being implemented and iterated first.
The core vitality of recommendation algorithms lies in data. Xia Wei launched a recommendation model adapted to short video scenarios a year ahead of time, accumulating user behavior data earlier.
Every time a user plays, stays on the page, likes, or favorites, they are "feeding" the algorithm, allowing the model to continuously optimize its accuracy.
This positive cycle of "more data, more accurate recommendations, higher user stickiness, and even more data" will create a snowball effect once it starts.
Even now, A.me, which doesn't even have characters yet, is unlikely to pose a threat to Douyin, even if it were to launch.
The reason is very simple:
Without Douyin, its competitors would be other short video apps, allowing it to learn and grow gradually.
After Xia Wei was reborn, its opponent became Douyin, giving it a powerful rival that had gained a first-mover advantage and a dominant market position.
Xia Wei's Douyin entered the market first. The two are in the same track, have the same core ideas, and are highly similar - Douyin is comparable to A.me's father.
Douyin has already firmly established itself at the top, holding a first-mover advantage and forming a competitive barrier;
As a latecomer, A.me could only follow the trend, lacking any differentiation, and was suppressed in every way...
You'll Also Like
-
I broke the dimensional barrier in Lord of Mysteries
Chapter 887 2 hours ago -
In Ultraman: Starting as Ultraman X
Chapter 146 2 hours ago -
It's normal for Pokémon to become gods, right?
Chapter 328 2 hours ago -
But I'm a male fox!
Chapter 368 2 hours ago -
To keep the heavens safe, I married the villainous boss.
Chapter 480 2 hours ago -
The villainous strategist's elite subordinates are all dangerous girls.
Chapter 292 2 hours ago -
One Piece: Kaido the Beasts is Playing House
Chapter 1165 2 hours ago -
Swallowing the Stars: Withdrawal Talent
Chapter 855 2 hours ago -
Douluo Continent: The Golden Tree of Martial Souls
Chapter 195 2 hours ago -
Ultraman: Who made this kid the human host?
Chapter 241 2 hours ago