In-depth analysis of Douyin e-commerce algorithm

Douyin e-commerce is like a game, and the algorithm is the rule of this game. This set of rules is valid for all users of the platform, whether they are content creators (merchants) or content consumers (users).

The purpose of the algorithm is to enable the platform to form a recyclable and healthy ecosystem while maximizing commercial monetization.

If we want to do e-commerce well on Douyin, we can only take advantage of the rules of the game and go with the flow after we understand the rules of the game.

Therefore, mastering the algorithm logic of Douyin e-commerce is the most basic and essential thing to play this game well. So how do we understand this mechanism?

Today, let’s talk about the Douyin e-commerce algorithm.

Let’s first look at the key word “weight”. What is weight?

Weight refers to the importance of a factor or indicator relative to a certain thing. It is different from the general proportion. It reflects not only the percentage of a factor or indicator, but emphasizes the relative importance of the factor or indicator, tending to the contribution or importance. Usually, weights can be judged and calculated by dividing indicators into multiple levels.

To put it simply, it is the evaluation that the system makes of each account, and this evaluation is measured through comprehensive data. The higher your rating, the higher the weight and the more traffic you will get.

In Douyin e-commerce, "weight" is mainly divided into: basic weight and real-time ranking weight.

1. Basic weight

As mentioned above, the basic weight is different for each account. The system will determine the weight level based on the comprehensive performance of the account within a certain period, and it is not fixed. If you don’t make progress, you will regress, and the fittest will survive.

For example, in the figure below, there are two different accounts. In case A, the first wave of streaming attracted thousands of viewers, while in case B, it attracted hundreds of viewers. It can be clearly seen that the two accounts are not at the same level. Therefore, the basic weight of the account determines the "volume" of the streaming.

It can be roughly divided into the following levels: Level E (cold start period, hundreds of viewers), Level D (thousands of viewers), Level C (tens of thousands of viewers), Level B (around one hundred thousand viewers), Level A (hundreds of thousands of viewers), and Level S (millions of viewers).

The above is about basic weights. But do you think you can sit back and sell goods without any worries just because your basic weight is high? nonono, there is another weight measurement mechanism.

2. Real-time ranking weight

We mentioned above that each account will have a basic weight. The amount of traffic after the broadcast is determined by the weight level, but the live broadcast room is like a reservoir. If there is no new traffic coming in, there will soon be no one in the live broadcast room. What is the distribution mechanism for subsequent traffic?

Tik Tok adopts a real-time horse racing mechanism, that is, after you start broadcasting and get a wave of traffic, if you still want traffic in the future, you need to compete with your competitors at the same level.

The system is ranking all the time, 5min, 30min, 60min, and the Douyin live broadcast traffic is constantly running a horse racing mechanism to screen out high-quality live broadcast rooms and allocate more traffic to them.

This picture basically clarifies the logic of the horse racing mechanism.

First, the initial push flow:

The initial traffic level is the basic weight we mentioned above. The higher the weight, the higher the initial traffic flow. At the same time, word-of-mouth points will also affect our traffic flow.

Enter data evaluation:

The system uses data to evaluate the performance of your live broadcast room every 5 minutes, 30 minutes, and 60 minutes, mainly based on interaction data and e-commerce data (specific key indicators are discussed separately below);

Performance on the sales list:

The system will then sort and compete at the same level. If you are higher than the previous one, you will enter the next level of traffic pool and get a new wave of traffic push, and then repeat the data evaluation. If you are lower than the next one, the traffic push will be reduced, or stopped, or even return to the initial traffic level.

Based on this mechanism, real-time ranking is continuously carried out to ensure the survival of the fittest. Therefore, only by planning and executing every detail of the live broadcast well can you beat your competitors and get more free traffic.

That’s all about weights. But when we do e-commerce, the premise of pursuing traffic scale is traffic accuracy. How does the system recommend precise groups of people to us?

Let’s first look at the key word “tag”. What is a tag?

Tags are symbols used by the platform to identify users. Every TikTok user will be labeled by the system, and the algorithm will recommend more accurate content to users; similarly, we creators will also be labeled by the system, and the algorithm will recommend accurate users to us based on the labels.

Tik Tok tags can be mainly classified into three categories:

1. Crowd tags

Divided into basic attribute tags and behavioral interest tags. The basic label is based on the user's registration information, mainly including gender, age, and region, such as: 18-25 years old + female + first-tier city. This combination is considered a population group.

In addition, the system will add specific labels based on the user's behavior trajectory on Douyin.

For example: If you like a mother and baby video on Douyin today, the system will think that you are interested in mother and baby related content, and will tag you with an interest tag. Next time, more similar content will be recommended to you. The same is true for behavior. For example, if you enter a mother and baby live broadcast room and eventually complete the shopping, you will be tagged as a mother and baby shopper. Behavior will be more accurate than interest.

2. Short video tags

There is a special labeling mechanism for short videos. For the short videos we post, the system will label the videos based on the effective viewing population, that is, the completion rate, the number of people who like and comment, as well as the product itself, and then match and distribute them to more potential interested users.

3. Live room label

It is divided into content tags and e-commerce tags, which are the same as the behavioral interest tags mentioned above, except that the above tags are for users, while here they are for creators.

For example, if users effectively watch, stay, comment, like, or become fans of your live broadcast room, the system will determine what kind of people are interested in your live broadcast room based on the data, and label your live broadcast room with content tags based on indicators such as watching, staying, etc. This is a shallow interest tag.

Then there are the e-commerce tags, which means when a user clicks on a shopping cart, a product, places an order for a product, or successfully completes a transaction in the live broadcast room, the system will label the live broadcast room with an accurate e-commerce tag. Subsequently, it will recommend products based on the existing tags or expand recommendations to more similar groups of people.

OK, now that we understand the labels, let's take a look at the relationship between labels and weights.

Tags are also divided into tags under basic weights and real-time tags. They are closely related to the weights mentioned above. Let's talk about them in detail.

1. Basic weight

Above we said that the basic weight determines the amount of traffic pushed by the system, so what is the relationship between the basic weight and the label?

The basic weight is formed by interest tags and e-commerce tags. To add interest tags to the live broadcast room, you only need to design the people and goods venue in the live broadcast room, as well as plan the script, to attract target users to watch the broadcast, stay, interact, and become fans to add the tag.

E-commerce tags require the accumulation of historical e-commerce orders in order to label the account with precise e-commerce tags. By achieving high-density transactions over a period of time at the beginning, the account can be labeled with a basic e-commerce population tag.

2. Real-time tags

This is easy to understand. Real-time traffic forms real-time tags. In every live broadcast, we must use precise product planning and paid traffic to continuously deepen account tags. The platform will explore the interactive and transaction groups in real time, and the traffic push model will become more and more accurate.

When we understand "weight" and "label", we need to pay attention to specific data indicators when putting them into practice. The platform direction has been adjusting. What are the key core indicators?

Before talking about specific indicators, we need to first understand the sources of traffic for the live broadcast room.

The above picture classifies the traffic sources of the live broadcast room, which are mainly divided into free traffic and paid traffic. We focus on the two main sources of free traffic: live broadcast recommendations and short videos.

1. Live broadcast recommendation-streaming data indicators

Live broadcast recommendation is the largest source of traffic for the live broadcast room. Let’s first look at the evaluation indicators of live broadcast recommendation.

Interaction data indicators: duration of stay, interaction rate (comments/likes, etc.), conversion rate/fan group addition rate, forwarding and sharing rate;

E-commerce data indicators: product click-through rate, transaction conversion rate, UV value, GPM (transactions per thousand impressions)

Others: account tags, weight, reputation score

The above data indicators will directly or indirectly determine the traffic of live broadcast recommendations. The system will not only evaluate a single data indicator, so all of them need to be done well, but there will be some key evaluation indicators, which are also what we need to focus on.

Since the rise of Douyin e-commerce last year, the platform has been constantly adjusting its algorithms as the ecosystem develops. In summary, it has made about three major adjustments, and the corresponding data indicators for key assessments are also changing.

In the first stage, key assessments include: duration of stay and interaction rate

Starting from April last year, the main data indicators assessed by the system were dwell time and interaction rate. For the platform, the focus of this stage was to cultivate user shopping habits. Keeping users was the most important thing. Therefore, the initial holding back of orders and flash sales directly hit the algorithm's vital point, making it easy to obtain natural traffic.

The second stage focuses on: UV value and transaction efficiency

The algorithm was adjusted in April this year, and the platform began to clean up a large number of low-quality live broadcast rooms and regulate merchants. For Douyin, the goal is to regulate the e-commerce ecological environment and further consolidate the commercialization chain, so the focus of the assessment is UV value and transaction efficiency.

The third stage, focus on assessment: GPM (thousand impressions per thousand sales)

This is also the stage we are currently experiencing. As the Douyin e-commerce ecosystem develops and takes shape, GPM will become a key indicator for platform assessment. The GPM indicator reflects the commercial value of the live broadcast room. To achieve this data indicator well, it is not as simple as focusing on a single dimension as we did before. Instead, it requires comprehensive improvement of comprehensive strength so that the platform will allocate traffic to you.

2. Short video-streaming data indicators

Short videos are the basic platform of Douyin, that is, content. We need to separate short videos from e-commerce. Similarly, the streaming algorithms of the two are different, and they also have different assessment data indicators.

The evaluation indicators for native videos are: completion rate, length of stay on the homepage, fan conversion rate, like rate, comment rate, and forwarding and sharing rate. Among them, the completion rate is the key indicator of the evaluation. You will find that many popular videos have a very high completion rate.

The evaluation indicators for e-commerce short videos are: live broadcast room entry rate, live broadcast room interaction & transaction data, stay time, interaction, and transaction.

Finally, let’s look at the relationship between the various traffic channels:

As shown in the figure above, the live broadcast room is the final landing scene. There are different sources of traffic, including paid traffic, short video traffic, live broadcast recommendation traffic, and fan traffic. Let's look at them one by one:

Paid traffic:

The display ranking is based on ECPM. Although you are paying for traffic, the higher the bid does not necessarily mean the more traffic you will get. The display of ads is based on this formula: Display = estimated click-through rate * estimated conversion rate * bid. The biggest role of paid traffic is to label the live broadcast room more accurately.

Short video traffic:

It is determined by data indicators such as the live broadcast room entry rate, live broadcast room stay, interaction, clicks, and transactions. At the same time, it will trigger a mutual heating mechanism. Short videos will accurately direct traffic to the live broadcast room, and the live broadcast room will reversely heat the short video, thereby obtaining more traffic.

Live broadcast recommendation:

It is determined by the indicators of stay time, interaction rate, conversion rate, fan conversion rate, product click-through rate, conversion rate, UV value, and GPM value, and is the largest traffic entrance for the live broadcast room.

Fan traffic:

The main data indicators are fan interaction and transaction feedback. Fans increase repeat purchases in the live broadcast room and create an atmosphere, thereby driving other data indicators.

To summarize:

1. Each traffic channel pushes traffic independently, and each channel has independent traffic data indicators, but users who enter the live broadcast room through different channels weight each other in terms of atmosphere, leading to a herd effect, thereby stimulating each other.

2. Paid traffic "stimulates" natural traffic to a certain extent. Paid traffic continuously adds accurate labels to the live broadcast room (basic & real-time labels), which can drive more accurate natural traffic groups with similar labels, but it does not stimulate "weight" (push traffic).

above. This analysis of the Douyin e-commerce algorithm explains in detail the principles of weight and traffic distribution. I hope it will be inspiring to you. Remember to like and share. Thank you for your support.

Author: A Tao and Chu Xin

Source: A Tao and Chu Xin