I analyzed 100 articles and summarized 5 truths about Toutiao's recommendation mechanism
|
This article has 2,970 words, and analyzes the relationship between data on reading conversion rate , reading progress, bounce rate, reading speed, interaction rate, and the amount of article recommendations . If you read it patiently, you will definitely gain something. ——Editor’s note In December 2016, the company decided to build a website about elderly care knowledge. After the website planning was approved and entered the development stage, there was no operation work for the time being. We planned to analyze user preferences/behaviors through self-media . Because we wanted to get user feedback quickly, we did not open a WeChat public account , but started with Toutiao . Toutiao passed the review on December 25, passed the novice period on January 3, opened the original tag on January 17, and opened the dual title function on February 17. In general, Toutiao is a self-media that can bring surprises to people. I say it brings surprises because the number of readers of some articles will always exceed your expectations, while the number of readers of some articles will make you extremely frustrated. Those who have operated Toutiao accounts know that the number of recommendations determines the number of readers of Toutiao articles. The recommendation mechanism of Toutiao really gives me a headache. A well-designed and proud article may have a very low readership, while a casually written article can easily get over 100,000 readers. According to the official statement, "what you care about is the headlines." Headline articles will be recommended to relevant users based on user preferences. In other words, label your article and label the user. When the labels are relevant, your article will be recommended to relevant users. Through my observation, I found that after each article is published, it will be recommended to 100-1000 users within 1 hour, and then the number of recommendations will gradually increase. Many articles that I think are good often die in the first hour. So, how can we please the headlines boss and get him to give us more recommendations? The official recommendation mechanism states that the amount of article recommendations is related to originality, conversion rate, interactivity, region, hot spots, etc. The official has explained it, so I won’t go into details. But what I want to say is that reading the official statement is like watching the news broadcast, do you understand? Okay, do you understand? Um? Do you understand? NO! In order to better improve the content quality of Toutiao, bring more valuable content to users, and improve user experience... I can't write any more, just to complete the KPI and appease the rich and handsome guy (the boss), I analyzed the articles published by Toutiao in the past few months, removed some unrepresentative data, and there were exactly 100 articles. How does Toutiao judge a good article? Toutiao provides data on recommendation volume, reading volume, comment volume, collection volume, forwarding volume, reading progress, bounce rate, reading speed, etc. I combine the comment volume, collection volume, and forwarding volume, which is called the interaction rate. According to the average value of each group of data, the top 20 recommended articles were selected for focused analysis. (The yellow part in the figure below is the data below the average) Through color comparison, we can see that among these data, the data with interaction rate not meeting the standard is the most. Therefore, I think the interaction rate has the lowest correlation with the recommendation volume, while the reading rate, reading progress, bounce rate, and reading speed have the greatest correlation with the recommendation volume. We all know that Toutiao uses machine algorithms to judge the quality of articles. Good articles are recommended to more users, while bad articles are stillborn. Through analysis, I found that what I originally thought as forwarding, collecting, and commenting are not the key criteria for Toutiao to judge the quality of an article. Why? Actually, I don’t know either. I guess it’s probably to prevent data brushing. After all, data such as forwarding and collecting can be faked, while other data can only be obtained from the behavior of the recommended users. How to increase the recommendation volume? Now that we know the key indicators that affect the recommendation volume, we can optimize the four indicators: reading rate, reading progress, bounce rate, and reading speed. 1. Improve reading rate The reading rate is actually the reading conversion rate, that is, the ratio of users who click on the article. Improving the reading rate is equivalent to writing a good title and choosing a good cover. But the question arises again, how to write a good title and choose a good cover? First of all, you should not use catchy headlines , because catchy headlines will cause problems with reading progress, bounce rate, and reading speed data. Secondly, be a moderate clickbait . The most original function of a title is to summarize the central idea of the article. A general summary may not have any highlights, so you can extract a specific point of the article for summary. In my operational work over the past few years, I have summarized a four-quadrant rule for title writing (as shown below), with useful, relevant, interesting, and curious as the four quadrants. The values gradually increase from the inside to the outside, and the titles are scored according to the four quadrant dimensions. The larger the area of the final graph, the better the title. The 4-quadrant rule for title writing analyzes the quality of an article through a graph composed of four-dimensional data. There are many specific evaluation criteria, which will not be elaborated here. The simplest principle for writing a good title is to "show the real thing". Don't let users choose articles like gambling on stones. Don't let users think. No matter how good or bad your title is, a bracelet of the same price will always be more attractive than a raw stone. 2. Improve reading progress It was also during this analysis that we discovered why carefully planned articles are often not favored, while casually written articles are "favored by the emperor". Carefully planned articles are often longer, while casually written articles are often finished in just a few words. The length of the article directly affects the reading progress. There are plenty of articles on Toutiao waiting for users to read, so why waste too much time on yours? Therefore, articles that can be read in three or five pages are the most popular, and articles that are too long should be left to one’s own devices. Short articles are the key to improving reading progress. Short articles also require us not to express too many opinions. Each article should only clearly state one issue. After all, everyone comes to Toutiao to see something new, and no one comes to attend class. 3. Reduce bounce rate The bounce rate refers to the ratio of users who jump out of the page before reading 20%. To reduce the bounce rate, we should not have too much nonsense at the beginning. The first paragraph should say what your title says. Writing a headline article is not like writing an official document. You don’t need to mention the background or significance at the beginning. For example, if you write "What kind of shoes are better for the elderly to choose for morning exercises?" in the title, first talk about the benefits and methods of morning exercises for the elderly, and then write about shoes at the end, then most users will jump out. Because the people who want to know how to choose shoes are those who are already doing morning exercises or preparing to do morning exercises. For such people, there is no need to explain the benefits and methods, just talk about shoes directly. Just like my article, which talks about the article recommendation mechanism, I started with a long-winded description of a lot of processes, which is actually unnecessary. Why write? Just show off. I added a quote so that you can see this sentence now. Adding a quote is actually like adding a burden, just like many articles now start with a sentence like "There are benefits at the end of the article", or put a more attractive cover picture (exposing a girl) at the end of the article. 4. Slow down your reading speed Regarding reading speed, it is difficult to find a specific and practical way to do it. Generally speaking, it is to improve the quality of content and increase readability, but this sentence is nonsense. Among the 100 articles we have published, the reading speed of "story-telling" articles is generally lower. Therefore, reading stories is a key to reducing reading speed. In addition, everyone knows that users have an "F-shaped" visual pattern when reading, that is, when reading an article quickly, the eyes appear in an F shape based on the time they stay in the reading. Therefore, to slow down the reading speed, you can divide the reading into more paragraphs, and each paragraph should not exceed one and a half lines in the Toutiao editing background. Write the key points of the content at the beginning of the paragraph so that users can grasp the key points and read the paragraph patiently. 5. Find the right keywords What I want to say here is the most important and also the most difficult to control point. The above 4 points are the key factors that determine whether an article can be recommended continuously, and keywords are the key indicators that determine how many users it can be recommended to. For example, our Toutiao account is about elderly care. Compared with news and jokes, it has much fewer users, probably less than one percent of them. If the recommended users are compared to a pond, the elderly care users are the small pond, the news users are the Yangtze River, and the users of joke articles are the ocean. The keywords of the article determine the size of the user pool. For an article with a large user pool, even if only one ten-thousandth of it is recommended, it is still more than one thousandth of an article with a small user pool. Therefore, when choosing an article topic, you should choose the right keywords. After analyzing 100 articles, I summarized the three characteristics of keywords: ▍Query the keyword index through Baidu Index. The higher the index, the larger the user pool; ▍Specify the keywords, such as the word "Spring Equinox". "Spring Equinox is not as good as Spring Equinox, and Spring Equinox is not as good as eating spinach. ▍Add keywords in the title, first paragraph, last paragraph, and picture annotations. The keywords should appear several times in the article. Toutiao accounts are recommended by machine algorithms, and algorithm analysis of keywords is not very user-friendly. Therefore, the machine's recognition can be improved through the location and frequency of keywords. I only realized after writing this that the article is too long. I think this article will only have a single-digit number of readers if it is posted on Toutiao. All the possible mistakes have been made. This is also the characteristic of the platform. WeChat public accounts , Toutiao accounts, Penguin accounts , Baijia accounts , etc. each self-media has its own characteristics , advantages, and pitfalls. There are many good things outside the door, but getting in is like a deep sea. Mobile application product promotion service: APP promotion service Qinggua Media advertising The author of this article @张小坏 is compiled and published by (Qinggua Media). Please indicate the author information and source when reprinting! Site Map |
Recommend
Ten thousand words long article analysis | How to create a trend through "category killer"
How to become a favorite concubine in palace dram...
Event review: How to run a successful event?
This activity was based on the WeChat service acc...
Douyin monetization 2020, quickly create a profitable Douyin
Whether it is operating a full-platform self-medi...
How to do Baidu bidding promotion? Learn these tips to quadruple your results!
I believe everyone is familiar with Baidu bidding...
Three typical APP promotion cases, which one would you choose?
Let’s talk about promotion again. It seems that t...
5 user segmentation models, have you used them?
When building a user operation system, the most i...
Brands must do these 5 things to stand out!
I believe that many people who do brand marketing...
How much does it cost to develop an audio and video mini program in Zunyi?
How much does it cost to develop the Zunyi Audio ...
How to use data to drive operational growth
Maybe we won’t get a satisfactory result immediat...
Introduction to Humanistic Psychotherapy
Introduction to Humanistic Psychotherapy A humani...
Mini Program advertising space is now online! Starting from 0.5 yuan per click, how to grab traffic bonus?
I finally waited for you, and luckily I didn’t gi...
How to quickly advance to the second level in search oCPC? Get it now!
How many conditions are needed to send search oCP...
How did the first batch of seed users of the product come from?
Many startup products are struggling to find the ...
Chen Changwen's method: Confessions of a community member
Every perceptive entrepreneur should practice soc...
How did the screenwriter of Nothing But Thirty respond to the controversy over the finale? Gu Jia believes in beauty
How did the screenwriter of Nothing But Thirty re...