Analysis of Toutiao’s recommendation mechanism!

Preface

If there is any company in China's current Internet industry that can be compared with Baidu and Tencent, ByteDance is undoubtedly it. In just a few years, this once very low-key company has rushed to the first echelon of the Internet industry.

Whether it is the former Neihan Duanzi, Huoshan Video, or Toutiao, Douyin, Kuaishou, Xigua Video, etc., any one of them is a high-traffic platform with users for several months, among which Toutiao is the core foundation of other APPs.

So how to gain a foothold in Toutiao, become a big V, and successfully realize monetization is a question that every online marketer is thinking about.

Many people may say that "content is king". Yes, content is the core that determines whether you can create a hit article. But I don't know if you have noticed that many times, even if you have written or filmed excellent works, and even have achieved great success by publishing them on other platforms, it will not work on Toutiao.

Why? Because each platform has its own recommendation mechanism, which is composed of a very complex comprehensive algorithm. It also examines your comprehensive data. Only when the comprehensive assessment indicators are met can it become a real hit.

So how does the recommendation mechanism work? Today we will explain in detail the working principles and operating principles of Toutiao’s recommendation mechanism, and have fun with Toutiao accounts with everyone.

1. Content Tags

Before a headline article is officially pushed, it needs to be reviewed by a machine first. After the content is targeted and featured based on the following indicators, the article is labeled.

How to determine article tags? Mainly rely on keywords. So how do we determine the keywords? That is the high-frequency words.

For example, in an article that introduces how to attract traffic and promote Toutiao, the high-frequency words are: "Today's Toutiao", "traffic diversion", "promotion", etc., then the system will use these three words as our tags. It is important to note that unconventional words should be avoided as much as possible, as they will increase the difficulty in understanding the article. For example, some substitute words for slang, such as "WeChat Overseas Half-Monthly Number", may only be understood by professional trading account merchants, and Toutiao's robots cannot recognize them.

2. User Analysis

The Toutiao recommendation system’s understanding of target users is obtained through big data analysis, and the judgment criteria mainly include the following three aspects.

(1) Basic information

Including user gender, age, region, commonly used apps, etc.

For example, a 25-year-old male user from Chongqing uses Toutiao to read entertainment news. The system will then try to recommend the types of articles that he likes to read to other users from the same region, gender, age, and who also like to use Toutiao. It will also learn from the reading habits what types of articles the user dislikes and reduce recommendations of this type.

(2) Focus on content

The content of attention is mainly determined from the perspectives of the accounts followed, channels followed, and topics followed. For example, if you follow a channel that talks about techniques for increasing followers on Toutiao, the system will try to recommend similar articles/accounts/channels/circles for traffic generation and promotion to you.

(3) Reading interest

Reading interests include: article types and keywords that users have read in the past, article types that similar users like to read, and article types marked as "not interested".

By understanding users' reading interests, we can further understand their reading habits. For example, if users like to click "like" when reading story-themed articles and click "dislike" when reading marketing-themed articles, the system will try to recommend story-themed articles to users and reduce or not recommend marketing-themed articles.

3. Recommendation Mechanism

Toutiao's content recommendations are not done overnight, nor are they static. Instead, they are recommended in batches. Douyin also has a similar recommendation mechanism. So how do they recommend?

1. Recommend in batches

There are two modes: one is multiple recommendations within a certain validity period; the other is recommendations with different validity periods (24 hours, 72 hours and one week).

2. First recommendation

The reading tags of the target users recommended for the first time have the highest matching degree with the article tags, and they are considered by the system as the users who are most likely to be interested in the article.

3. The first reading data determines the second recommendation amount

The user reading data of the first recommendation, including click-through rate, number of collections, number of comments, number of reposts, completion rate and page dwell time, determines the second recommendation volume. If the comprehensive index of the first recommendation is relatively high, it means that the article is suitable for this type of user. Then the system will increase its recommendation efforts the second time and recommend it to more people, and vice versa.

So, many people find that no one reads the content they post right away, or that it goes viral in just a few hours. This is based on this recommendation mechanism.

4. Several factors for not getting recommended

(1) Content is not vertical

When we register for a Toutiao account, we need to choose a personal field. Even if we do not post according to the field, the system will determine the field according to the content and keywords we frequently send. If the new content you send does not match the field, the system will re-identify it and then recommend it. This will prevent the article from being recommended to the corresponding users in a timely manner, and naturally it will not get a high recommendation volume and playback volume.

(2) Article content

The content of an article actually does not only refer to the main text of the article, but also includes the article cover, pictures, labels, title, etc. If the title and cover are not attractive enough to users or the image quality is too low, the accompanying pictures are not strongly related to the content, and the article content is of poor quality, etc., the article will have a low click-through rate.

(3) The user base in the selected field is relatively small.

The user groups in some fields are very niche, and there are fewer readers on Toutiao. Therefore, even if the quality of the articles we write is high, it is difficult to attract a large number of readers as quickly as in large fields with high attention, such as the entertainment field and the automotive field. For example, in fields like rock and mechanics, it is difficult to get a large number of recommendations on Toutiao, after all, they are not very compatible with Toutiao users.

(4) There is too much similar content.

Of course, choosing a field that is too popular is not necessarily a good thing, because there are too many competitors and many strong ones. As long as a hot topic emerges, many big Vs will start posting, but the number of users paying attention is limited. On the one hand, you will definitely be at a disadvantage when competing with those professionals. Moreover, as more people write about hot topics, they will no longer be fresh, and correspondingly fewer people will read them.

( 5) Non-original content may be de-duplicated, resulting in low recommendation volume, such as when an article is published multiple times.

(6) The content is short-lived, which results in a short recommendation lifespan. For example, hot events usually fade away after a few days, and the recommendations will naturally disappear quickly.

The above lists a lot of content about Toutiao’s recommendation mechanism. In summary, if you want to operate Toutiao for a long time and achieve good results, you must continue to output in the vertical field and constantly test and explore in the process of practice. I believe that you will soon be able to find a set of creative methods that suit you for high recommendations and high playback volumes.

Author: Bu Jingyun

Source: Drainage Xiabu Jingyun