Tiktok de centralized recommendation system

1. What is the algorithm for tiktok? TikTok Technology Co., Ltd. Tianjin

simply put forward: the algorithm of shaking is actually a funnel mechanism, which is basically consistent with the principle of recommendation algorithm of the headline's de centralization. Tiktok / BR > it has three tiktok steps:
first, cold start flow cell exposure
assume that 1 million people are uploading short videos every day on the jitter, and the jitter will randomly assign each cold video to a cold start flow pool with average exposure. For example, after each short video is released through audit, there are an average of 1000 exposures
Second, data selection
tiktok will analyze 1000 points of exposure from 1 million short videos, analyze the data of points, attention, comment, forwarding and so on, and pick out videos with more than 10% indicators, and then distribute 100 thousand times on average. Then go to see which are like, follow, forward and comment more than 10%, and roll into the next round of larger flow pool for recommendation
Third, the boutique recommendation pool
through round after round of verification, short videos with extremely high like rate, playback completion rate, comment interaction rate and other indicators are screened out to have the opportunity to enter the boutique recommendation pool. When users open the pool, they can see videos with tens of millions of likes
all the dry goods and skills shared next are closely around the core point: by increasing the number of likes, attention, comments, forwarding rate and other indicators, we can get more accurate official recommendations and win more exposure.

2. Tiktok is a de centralized algorithm platform, which means that any account has the opportunity to have millions of fans and even millions of fans. Even if we don't have any traffic, as long as our content is popular, more and more people will pay attention to it. So if you do the content in the right way, you can earn the first barrel of gold in the tiktok. If you don't know anything, please ask me in private

3. 1-5000-100: that is to say, you'd better play more than 5000 videos within one hour, and if you like more than 100 videos, you're more likely to be recommended by the system
four dimensions: the number of finished broadcasts, the rate of likes, the number of forwards, and the number of comments and interactions
publishing time: it's best to publish at 11-13 noon, 5-7 PM, and 9-11 PM. These three periods are the peak periods of user access. After your video is published, the system will basically recommend some traffic.

4.

1、 Recommendation principle of intelligent algorithm

the essence of intelligent algorithm recommendation is to match the most interesting content for current users from an aggregate content pool

this content pool contains tens of millions of content every day, covering 15s short video, 1min long video and 5min super long video

when matching content to users, the platform mainly depends on three factors: content, users and their interest in content

How does the

system understand our creative content

when describing the content, the platform will mainly rely on keyword recognition technology: by extracting the keywords in writing and video, the content will be roughly classified according to the keywords, and then the classification will be refined according to the keywords in the subdivision field

for example, the keywords of video and content are "Ronaldo, football, World Cup"

most of the key words belong to sports vocabulary. You will first classify your works into sports categories, and then subdivide them into "football", "international football" and other secondary and tertiary categories according to the specific key words

user characterization

through this series of comparison and analysis, the system speculates and restores the basic attributes of a user, for example: TA may be a male who is traveling and likes football, car and other categories

the system will classify the above user characteristics as the user's tag

User tags are mainly divided into three categories:

1) basic information of users (age, gender, region)

2) user's behavior information (attention account number, history wandering record, like collection content, music, topic)

3) reading interest (reading behavior, user clustering, user tagging)

according to the user's information and behavior, the system analyzes and calculates the user's classification, topics, people and other information, so as to complete the system's characterization of users

the essence of recommendation algorithm

uses the characteristics of works (keywords, labels, popularity, forwarding, timeliness, similarity), user preferences (short-term click behavior, interest, occupation, age, gender, etc.), and environmental factors (region, time, weather, network environment) to fit a function of user satisfaction with content, It will estimate the user's click probability for each work, and then rank all the works according to their interests from tens of millions of content flow pools in the system. The top 10 works will stand out at this time and be recommended to the user's mobile phone for display

it's probably like this. If you want to learn, you can write to Xiaobian in private

5. Tiktok algorithm is actually a funnel mechanism, which is basically consistent with the principle of recommendation algorithm of the headline's de centralization. Tiktok tiktok is divided into three steps:

first, cold start flow cell exposure

suppose that 1 million people upload short videos every day on the shaking, and the jitter will randomly assign each cold video to a cold start flow pool with an average exposure. For example, after each short video is sent out by auditing, there are 1000 exposures

second on average. Data selection

tiktok will analyze 1000 points of the 1 million short videos, analyze the data of each dimension such as point, attention, comment, forwarding and so on, and then pick out videos with more than 10% indexes, and then distribute 100 thousand times on average. Then go to see which are like, follow, forward and comment more than 10%, and roll into the next round of larger flow pool for recommendation

thirdly, the boutique recommendation pool

through round after round of verification, short videos with extremely high liking rate, playback completion rate, comment interaction rate and other indicators are screened out to have the opportunity to enter the boutique recommendation pool. When users open the pool, they see videos with tens of millions of likes< br />
: " We are paladins. We can't let revenge occupy our consciousness& quot;,

6. The hardware is completely OK, and the graphics card 1060 can be fully configured. The processor i78750h is all right. It's smooth

7.

bus line: Metro Line 4 → Metro Line 7 → Metro Line 1, the whole journey is about 31.9 km

1. Walk about 620 meters from Wenjiang District to Guanghua Park Station

2. Take Metro Line 4, pass 9 stations, and reach Cultural Palace Station

3. Take Metro Line 7, pass 7 stations, and reach South Railway Station

4. Walk about 70 meters, and transfer to Metro Line 1

5 Take Metro Line 1, pass 4 stops to Jincheng Square Station

6, walk about 380 meters to Shuangliu global center

8. Of course, Shuangliu is more suitable to buy a house. Shuangliu belongs to the South Gate of Cheng. Now the south gate is developing very well, but the house price here is much higher than that in Wenjiang