Rebirth 16: From Hijacking Douyin to Dominating the World
027 Low-key School Beauty Wu Xiaoman
In March, spring arrives, all things revive, and it's mating season for animals.
Half a month later, the spring breeze on the street has lost its biting chill and become gentle.
The girls on the Communication University campus have already changed into light spring clothes, and their flesh-colored tights are shining with a captivating radiance, adding a touch of vibrant spring.
Rao Jia's arrival put the butterfly effect on a fast track of development, and also gave Xia Wei some free time to enjoy the beautiful spring scenery.
Algorithm engineers, video engineers, finance professionals, operations staff, content creators... one new employee after another was recruited and hired by Rao Jia.
It's worth mentioning the newly hired algorithm architect:
His name is Ding Ming, and his monthly base salary is 55K, making him the highest-paid employee in the entire company.
He is 36 years old this year. He has a bachelor’s and master’s degree in computer science from East China University of Science and Technology. When he left the company, he was an algorithm architect for character processing, with a job level of 2-3.
He is one of the most experienced engineers in character processing and best at online implementation.
Proficient in using distributed machine learning platforms, deeply involved in the iteration of Toutiao's news recommendation algorithm, and capable of independently leading a team to implement a recommendation system.
He left the character after losing to someone less capable than him in the advancement round, and was found by Rao Jia.
Xia Wei had a long talk with him, outlining his idea of combining "machine learning platforms" and "recommendation algorithms" into personalized short video recommendation algorithms.
After hearing this plan, Ding Ming was amazed by Xia Wei's talent.
He not only understands ByteDance's stuff, but also pioneered the idea of adapting recommendation algorithms from "text and images" to "short video" scenarios.
This idea is very constructive and seems very promising; no company on the market is doing this—not even a single character.
He was very interested, and they immediately hit it off, with him agreeing to become the algorithm lead for The Butterfly Effect.
Once he joins the company, we can discuss more in-depth technical solutions, which is not convenient to talk about during the interview.
Xia Wei directly provided the first version of the algorithm idea and asked him to implement it. The specific steps were very simple:
Step 1: Decouple ByteDance's core modules of "information recommendation algorithm" and "distributed machine learning platform" while keeping the underlying architecture unchanged, saving 90% of the underlying development cost;
Step 2: Reconstruct the content understanding module to break the limitations of text and image scenarios;
Instead of relying on text keywords and paragraph semantics, the original information algorithm extracts core action features from short videos.
Then, by using image recognition, audio analysis, extraction of video cover features, background music style, and scene labels, the scene adaptation on the content side is completed.
The third step is to build a personalized interest recommendation and filtering system to accurately filter out videos that users may like.
On the one hand, by analyzing the videos that users have previously viewed, liked, and lingered on, we can deduce the content they have a long-term preference for;
On the other hand, it captures users' current video-watching behavior and updates their short-term interests.
This allows for both long-term and real-time interest capture, accurately matching each user's tastes.
Step 4: Combine user interest tags and short video content features with short video-specific behavioral data—such as three-second retention, completion rate, and like-to-view ratio.
Input a shallow DNN model, perform deep learning and retraining, and continuously optimize the interest matching weights.
This allows for the integration of the two technologies, enabling the implementation of the world's first personalized short video interest recommendation algorithm.
— This enables personalized content delivery tailored to each individual user.
Ding Ming froze on the spot, his small hands trembling, his eyes filled with disbelief, and his mind buzzing.
He has been deeply involved in algorithms for seven years, from Tencent to ByteDance, and personally participated in the iteration of Toutiao's news recommendation algorithm.
He knew better than anyone that the hardest part of adapting text and image recommendations to short videos was the creativity—being the first to come up with the idea.
The second point is how to build a new feature system, content understanding, and real-time modeling of user dynamic interests suitable for short videos.
Xia Wei provided a complete end-to-end engineering solution, clearly outlining every step, and offering practical solutions that could be implemented immediately.
Although there isn't a single line of code, a rudimentary short video interest recommendation algorithm can be implemented simply by following the instructions.
Once this solution is implemented, it will most likely be a game-changer, overwhelming all current short video platforms.
Forget about big shots, Meipai, and Miaopai, all of them can get lost.
His judgment was correct. Xia Wei understood the brilliance and power of this plan better than he did, and couldn't wait to see the effect of it going online more than half a year ahead of schedule.
Tell him directly:
The first version doesn't need to be perfect, just enough to work and make sure it goes live.
After launch, iterate and optimize gradually.
The algorithm requires a large amount of user data for deep learning; the more data, the more accurate the recommendations.
Go live quickly to acquire a large amount of user data, and evolve and iterate as it is used.
Ding Mingshen agreed wholeheartedly and followed his advice without question.
Working with two newly recruited experts, we're putting in late nights every day to launch the world's first short video recommendation algorithm as soon as possible...
After a rapid build-up at the end of February, and the addition of key personnel, the company's framework was completed.
All departments are working together, the content department is making rapid progress, algorithm development is progressing at a fast pace, and the products are changing every day.
For most of the past two weeks in March:
At Xingdong Entertainment, Chen Mo's team is working diligently to modify the "Huangka APP" according to his plan, and it will be launched soon.
The campus content team—Content Department 1—gradually expanded to 50 people, and a content planner named Li Shuyi joined as Zhang Ting's assistant.
Three student group leaders were promoted, leading the entire team to produce an average of 400 short videos per day.
With the help of Teacher Wang from the school's entrepreneurship incubation center, the school quickly approved Xia Wei's entrepreneurship project and granted him a special 120-square-meter shared office space.
The office is located in Longchuan Cultural and Creative Park, 1.5 kilometers away from the school. It has shared meeting rooms and is rent-free for two years.
The content team works here, as does the second department. The only full-time employee is Fan Xuefei.
By a twist of fate, my roommate Lu Zicheng was transferred to the Content Department II to work as a PR agent—oh, no, I mean operations manager.
He, along with two other part-time students, formed an operations team, with him serving as the team leader.
The three people registered 100 accounts on five platforms: Weibo Music, Douban Music, Tencent Music, Zhihu, and Tieba.
They stirred up trouble, manipulated the wind and rain, and guided music lovers to appreciate "Drop Tower" and "Wild Wolf Disco".
Lu Zicheng gradually grew to love the feeling of being the mastermind behind the scenes, and his online marketing skills became increasingly proficient.
He mastered the method of guiding public opinion that Xia Wei taught him.
My roommate, Fatty Wang, became the leader of the music group in Content Department 1, and he also has two part-time music students under him.
The three of them were responsible for preparing the content for the music account.
My third roommate, Li Zhuo, is a second-generation rich kid and is indeed spoiled.
Unable to adapt to the life of attending classes every day while also working part-time, he made excuses not to do part-time work and returned to the regular routine of attending classes, playing games, and dating.
Fan Xuefei, the manager of the content department, after securing Kang Ninghua, successfully secured Zhao Lusi, who was studying at Wanbei Mingdao University.
She failed to make the cut in the 2016 Super Girl Wanbei District auditions, but she was determined to become an idol. Fan Xuefei took advantage of her vulnerability and won her over with a trip to Wanbei.
She's still studying there, and the company will remotely manage her development; she only needs one opportunity to become a sensation.
Another potential signing, Bai Lu, a model working in Shanghai, has seen her contract negotiations stalled for the time being.
She neither refused nor agreed yet, saying she wanted to wait and see—to see if the company could make a few people famous, and then decide after seeing a role model.
After Rao Jia joined the company, Xia Wei's main job became being an interviewer.
She was responsible for searching, screening, and conducting the first and second rounds of interviews, while Xia Wei was only responsible for the final interview.
Aside from interviews, Xia Wei devoted most of his energy to Content Division 1.
The second content division focuses on incubating top-tier celebrities and online influencers.
Liu Yuning, Kang Ninghua, and Zhao Lusi are the first three; there will be many more after them.
The Content Division is responsible for creating content for "non-real-person on-camera accounts".
It covers multiple categories including music, lifestyle, science, humanities, and emotions.
We will prepare tens of thousands of short videos in advance and categorize them carefully.
When Douyin was launched, the official team created hundreds of accounts that did not feature real people on camera and uploaded tens of thousands of short videos.
These accounts are "initiators," they are "shills."
As the number of registered users on the platform increases and the amount of content uploaded by users increases, a virtuous cycle and self-sufficiency are achieved, and the platform can then retire.
Xia Wei recruited people well in advance and developed his own content in preparation for the app's launch...
As the work proceeded smoothly, it was noon on March 23rd.
It has been exactly one month since Xia Wei was reborn.
He rode his Merida Duke 600, a legendary motorcycle for the Sichuan-Tibet route, and carried a Lenovo Legion motorcycle on his back as he left the school through the west gate.
I cycled 1.8 kilometers along the auxiliary road of Chaoyang Road and arrived at Building 3 of the Cultural and Creative Park in the West Campus.
The West Campus Cultural and Creative Park serves as an industry-academia-research integration incubation platform, a student entrepreneurship practice platform, and a youth entrepreneurship incubation base in Chaoyang District for Longguo Communication University.
We focus on media-related projects and provide full-cycle incubation services for student startups.
There are shared printing rooms and meeting rooms, and even shared camera equipment lockers and editing workstations. You can use them by paying a deposit and rent.
It's only 1.5 kilometers from here to the main campus, and there are regular school buses running between the two locations.
The fastest way to get around is by bicycle, with shared bikes available everywhere, taking only 5 minutes; walking takes only 20 minutes.
The school allocated Xia Wei a full 120 square meters of shared office space, located on the first floor of Building 3.
There are 500 square meters of shared office space here, which adopts the layout of "open workstations + independent partitions + shared functional areas".
There are 16 startup teams working here.
The entire southeast corner is the domain of the "butterfly effect".
Xia Wei arrived during lunch break. The huge office area was almost empty, with only a handful of people and extremely quiet.
Only Wu Xiaoman, the "most low-key school beauty," was looking at her computer.
She was extremely beautiful, with a subtle and enduring beauty hidden beneath her simple clothing.
Her skin was fair and delicate, and her eyebrows and eyes were exquisite, but she always kept her round face down and wore large black-rimmed glasses.
He was dressed simply, and his whole being was enveloped in a somber aura that kept strangers at bay.
A simple black sweater couldn't hide her amazing curves, outlining her full and proud figure.
Spring has arrived, and the warmth is growing stronger. "She" can no longer hide herself, like a butterfly emerging from its cocoon, displaying her proud springtime beauty.
She is really beautiful; her features and figure are absolutely stunning.
However, most of her beautiful features were obscured by her large glasses, making it impossible to discern her power.
On the contrary, Ultraman Tiga is full of energy and unleashes its full power.
If someone walks by, their attention will most likely not be drawn to her face, but rather to her upper body...
Wearing headphones, she sat in the corner cubicle, intently watching the screen, unaware that Xia Wei had come to stand behind her.
The screen displays vibrantly colored panels from a Korean comic.
The male and female protagonists are in intimate poses, their clothes are half off, and ambiguous light and shadow fill the entire frame.
This isn't an erotic comic; it's a large-scale romance comic, comparable to a movie, "more conservative than Category III films but more open than regular movies."
Xia Wei remained silent, resting his chin on his hand and leaning against the partition, a meaningful, wicked smile playing on his lips.
Normally, Wu Xiaoman is sullen and bookish, but at this moment she is all smiles.
A vibrant energy, rarely seen in everyday life, is hidden in the blink of an eye.
After watching for a while, she reached for the snacks, but caught a glimpse of Xia Wei out of the corner of her eye. Her fingers jerked back, and her hand holding the mouse tightened instantly.
She turned around abruptly and met Xia Wei's smiling eyes. She was struck dumb, frozen instantly, her whole body motionless.
Only the full headlights trembled slightly, telling of their owner's shyness and panic.
He glanced at the computer screen unconsciously, then looked up at Xia Wei.
A blush crept up her cheeks at a visible speed, quickly spreading to the tips of her ears, and even her neck was flushed with a thin layer of red...
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