Taking Jianshu as an example, let’s explore how UGC content products are cold started!

What is a cold start and how to do it? Is there a specific case to illustrate it? Today, taking Jianshu as an example, let’s explore how UGC content products do cold start and what their content operation mechanism is like.

1. Cold start strategy

1. Data Collection

1. Data collection methods

When reviewing a product's cold start strategy, the most important thing is to collect the initial data information. Jianshu started its public beta on April 13, 2013, and it started on the PC side. Therefore, if I want to study its cold start strategy, I have to find the content information of that time period. I will use Baidu search engine to collect the information.

Before deciding to use Baidu search engine as a tool for collecting information, one problem needs to be solved first: whether the data obtained by Baidu search engine is feasible, that is, accurate. There are two aspects to this question. The first aspect is whether the Baidu search engine itself has the ability to accurately search data. As the largest search engine in China, the data collected from the Baidu search engine can be said to be relatively accurate. The second aspect is whether the data collected by Baidu can represent the data when Jianshu is cold started. I think it can represent, why? Because the main criteria for search engines to provide information to users are relevance and importance (high content quality), the criteria for the URL pages they include are also these two. As a content-based platform, Jianshu must provide users with high-quality content to complete the cold start. Therefore, if the important content during the cold start period is included, it must be included by Baidu. If it is not included, then the content quality is not very high. The role of these contents with poor quality in the cold start of a content-based product can be ignored. Therefore, the cold start information obtained through Baidu search engine is feasible.

2. Specific operation methods

Open the Baidu search engine. If I want to find all the articles on Jianshu in a certain period of time (for example, from April 13, 2013 to July 13, 2013), I can do the following:

Enter the keyword "Jianshu" in the search box, select the time from April 13, 2013 to July 13, 2013 in the search tool, choose to search all web pages and files to avoid omissions, then choose to search on the Jianshu website (www.jianshu.com) to filter out redundant information, and click OK, as shown below.

In this way, we obtained all the articles published on Jianshu from April 13, 2013 to July 13, 2013, and then screened and adopted them according to needs.

2. Determination of the cold start phase

For a UGC content product like Jianshu, what is a cold start? It can be simply understood as the initialization of content. To put it simply, the product is out, but there is nothing there, so let’s see what to do. So, can we determine at which stage a product’s cold start is? For products like Jianshu, product initialization definitely requires filling it with content. So, is it possible to determine the cold start phase based on the amount of content produced by the product within one year since the product was released (the cold start time of a product should not exceed one year, otherwise it will be really cold)? I used the search method mentioned above to search for the amount of content published on Jianshu every month from April 13, 2013 to April 13, 2014, and made the following chart.

As can be seen from the above figure, in the first three months, the number of articles published by Jianshu was relatively high each month, with an average of 228 articles published each month. However, starting from the fourth month, the number of articles dropped sharply, from 309 in the third month to 47, and then until the ninth month, it remained at around 70. From the tenth month onwards, a faster growth trend began to emerge.

Now the question is, why did the release volume remain very high in the first three months, but suddenly drop sharply in the fourth month, followed by a period of stability and then growth again? The only explanation is that the first three months were the cold start period of the product, during which the operations staff carried out the content filling phase. From the fourth month onwards, the content on the product began to be produced by users themselves, and subsequently increased through operational means (I feel that the reasons behind the sharp increase starting in November are worth analyzing, but due to the limitations of the analysis topic, they will not be involved).

To sum up, I tentatively set the cold start time as the first three months, that is, April 13, 2013 to July 13, 2013. In addition, taking a step back, if the above argument is wrong, then three months of cold start time is sufficient for cold start analysis of a product.

Now that the time period for the cold start has been determined, do we need to organize all the data from the first three months? I don’t think it’s necessary. In fact, the cold start in a strict sense should be the first month. The following two months can be said to be a continuation and repetition of the cold start method, so that the product content can be rapidly expanded, the content can be rapidly grown, and users can be attracted. By collating and analyzing the data from the first month, we can have a clear understanding of the product's entry point, target population, and specific operating methods.

3. Entry point

Through the search method mentioned above, I found the data from the first month (April 13, 2013 to May 13, 2013) for observation and organization.

During this month, Jianshu users published a total of 210 articles. In general, they can be divided into four categories:

• The first category: platform usage guides published by Jianshu and introductory articles that highlight product positioning. This section contains about 20 articles in total, which address the question of how new users should use the platform and what the platform does.
• The second category: technical articles, including code writing and framework building. This type of article accounts for about 30%.
• The third category: The third category is humanities articles, including film reviews, book reviews, and essays. Compared with the current Jianshu articles, this type of article is obviously more in-depth. This type of articles accounts for a relatively large proportion, about 60%.
• The fourth category: Activity submission category. Needless to say, there is only one article, which is an essay competition officially organized by Jianshu.

From this, we can conclude that during the cold start period, Jianshu chose two entry points: one was technical topics involving programming codes, and the other was articles that expressed personal opinions and emotions (relatively broad).

Jianshu’s entry point is different from Zhihu. Zhihu’s initial entry point is the Internet. Users discuss topics around the Internet and then expand. Compared with Zhihu, it is obvious that Jianshu’s choice of topics is broader and not very clear. I think there are three reasons for this phenomenon:

1. First, it covers a wide range of user groups, and the breadth of the subject matter can expand the user range. In addition, although the scope is relatively broad, there are more targeted and highly targeted programming articles within the broader scope. This approach can quickly attract a group of users.
2. Second, Jianshu’s main business in the early days was as a writing tool, and these broad contents may just be an external manifestation to prove the applicability of the writing tool.
3. Third, the content filled in this stage is of a testing nature, testing user reactions and then adjusting the content based on user feedback. Broader topics can reduce the time cost of finding user interests and help with product positioning.

4. Target population

Based on the thematic analysis of the above product content, we can conclude that Jianshu’s initial target audience was a group of people who wanted to write and share, among whom there was a group with a clear focus, namely programmers .

By reading specific articles, we will find a phenomenon that the Jianshu user group does not need to write articles with strict logic and depth (except technical articles) like Zhihu users. They only need to record their own essays, write their own thoughts, express their emotions, and express their opinions, that's all. Therefore, we can further profile the target group, that is, a group of people whose writing ability and professional level (except technical articles, because I can't understand them) are not very high but who want to write and share.

(V) Specific operation methods

So how does content initialization work? I think there are three main ways.

First, publishing articles through Jianshu’s official account is mainly done by yourself. For example, articles introducing the platform are operated in this way, and there is nothing much to say about the others.

The second method is to invite seed users to write on the platform. Through personal relationships and employee relationships, we first spread the message in a small circle to attract seed users and enable them to continue producing content on the platform. For example, the author Zhao Yunbo was quite active in the early stages of the product launch and produced a high volume of articles. And now, he is still active on Jianshu, and the last article he updated was on July 11. Of course, there must be more than one such seed user. For example, there is an account called "Tashue Wuhen" (Taxue Wuhen), and if you want to dig deeper, there will be a lot of information. However, we are only exploring its operation methods, so we will only give one example and will not go into details about the others.

The third type is to officially register multiple accounts to produce content on the platform, which is often called a vest account. This can be said to be the majority. When Jianshu was cold-started, most of the content was produced by itself. This can be inferred from the total number of words written by its authors. The total number of words written by many accounts is between 6,000 and 14,000. Obviously, this type of account has not written again after writing 1 to 3 articles. Moreover, this cannot be the loss of users, because it is a cold start, and the product has no ready-made users at this time, unless they are seed users. However, if the seed users are managed to leave after writing two or three articles, it is time to reflect on this.

In fact, at this point, the basic operation methods of cold start have been basically analyzed, but there is still an important link missing. That is to maintain close contact with users (real new users), interact, generate feedback, and form a closed loop. However, I did not find such information. After spending a lot of effort, I finally found the legendary response from the CEO of Jianshu to a user. As shown below.

Although there is only this one item, it shows that the official is doing this to maintain close contact with users. So why is there only this one item? This may be circumstantial evidence for the third operation method mentioned above.

In short, it is important to establish a user maintenance mechanism during the cold start process, so that if a new user arrives, he must be developed into a seed user to accumulate foundation for the product.

2. Content Operation Mechanism

1. What is the content of Jianshu?

What is content? Information that can attract users, make them stay and invest time is content, but this is a general concept. For a specific product, the content of the product can be concretized. So for Jianshu, what is its content?

In fact, Jianshu, as a UGC community + social (light social) product, has built a platform for users to contact each other. The content on Jianshu is all generated by users, and its content is articles (including pictures in the articles).

So the question is, with so many users producing content, how are these contents organized and operated? How do we connect readers and authors to realize their respective values?

2. How is the content presented?

When talking about the presentation of content, we must include the first impression of the content: the interface. Just like the first impression of a person is his appearance and clothes (for example, many people see me and only say two words: so handsome), the interface of Jianshu gives people a feeling of freshness and being relatively well-behaved. I think this has a lot to do with the fact that most of its users are young people. We will just briefly mention the interface here and not go into details. We plan to focus on how Jianshu presents articles produced by many users, that is, what means it uses to present them. (As for the basis for presentation, it will be analyzed in the control method.) In fact, from a broad perspective, Jianshu presents two types of content to users: articles and special topics.

1. Article

An article is a separate article written by the user. The left two-thirds of each article displays the title, author, publishing time, number of reads, number of comments, number of likes, topic source, and number of rewards. The right one-third is the first picture of the article.

There are the following 8 sub-sections under the article section. From left to right, they are Popular, New on the List, Daily Report, Seven-Day Popular, Thirty-Day Popular, Prize-giving Events, Jianshu Publishing, and Jianshu Podcast.

Among them, "Hot" is the default homepage of Jianshu. When you open the Jianshu app, what is directly presented to the user is Hot. The Hot section consists of two parts: homepage and popular articles.

Popular articles are listed below the homepage, which takes up a quarter of the entire page and is in the form of a picture with the title and introduction of the article displayed above it. The articles below the homepage are arranged from top to bottom according to popularity.

The other columns such as New List, Daily, Seven-Day Hot, Thirty-Day Hot are arranged in the same format as ordinary articles, so I will not go into details here.

2. Special Topics

Under the topic section, there are two sub-sections: popular and recommended. The page has different topic pages. Topics are a feature of Jianshu. Each user can enter a topic and read the articles under it. They can also create a topic. At the same time, if users can follow the topics they are interested in, they can choose whether to receive push notifications for updated articles in the topics they follow.

3. Content Control Methods and Quality Standards

On Jianshu, users can edit and publish any article on it, so it seems that there is no need for quality standards for articles on Jianshu. However, as a platform that accurately matches content with users, how is this matching process achieved? For example, the articles on the homepage. The mobile phone screen is so small, so what is the basis for sorting so many articles? Therefore, we are about to analyze the content: the way of controlling content presentation. Jianshu's content control itself contains quality standards. Quality standards are implemented in the process of controlling content and matching content with users. There are two ways to control Jianshu’s content: machine algorithm and manual editing. The standards for different control methods are different. Next, I will analyze the implementation principles and standards of these two different control methods respectively.

1. Manual editing

(1) Home page review

Everyone can contribute to the homepage. If accepted by the homepage editor, the accepted article will appear on the homepage and will be recommended on the homepage.

Standard: It can be said that the standard is the editor's judgment. We don't know whether Jianshu has a set of evaluation standards internally. But in general, the following standards can be roughly summarized: there is no limit on the type, but there will be relatively high requirements in terms of professionalism, readability and depth.

(2) Manual pinning

According to an article published by Liu Miao, the operations director of Jianshu, on the webmaster's home page, under the control of machine algorithms (which will be introduced below), some good articles (in-depth and good articles) are not ranked on the homepage for various reasons (the most widely circulated articles are not necessarily good articles, and articles with a high number of readers are not necessarily good articles). Therefore, the editor of Jianshu will have the right to directly recommend the article to the homepage to allow good articles to be exposed and not remain silent. As for the selection criteria, they are the same as those for recommended articles on the homepage.

(3) Manual recommendation

In fact, manual recommendation and homepage review should belong to the same category, both of which rely on human judgment to select content. However, due to the particularity of the homepage, I want to separate the two to maintain the uniqueness of the homepage. The four aforementioned prize-giving activities, Jianshu Publishing, and Jianshu Podcasts are all manual recommendation methods. Choose high-quality content and update it over time. If there is content today, update it. If there is content tomorrow, update it tomorrow. Next, I intend to analyze its main contents and quality standards one by one.

① Prize-giving activities: This section is equivalent to a column for the official activities of Jianshu. If Jianshu has any activities, they will be announced here. The standard is the standard of the activity itself.

② Jianshu Publishing: It is a platform for Jianshu authors to publish their own contents in the form of e-books. The contents of this section are introductions to authors and books that have published e-books through Jianshu Publishing.

The presentation method has fixed standards, which basically consists of 10 parts: introducing yourself, how you know Jianshu, why you write on Jianshu, your story with Jianshu, your books, what motivated you to finish writing a book, the biggest gains and difficulties in writing a book, reading habits, writing habits, and a message to Jianshu and its readers. The overall tone requires a more emotional expression.

③Jianshu Podcast: This is a relatively special content. It is a new feature launched by Jianshu on April 28 last year. Users can contribute to the Jianshu blog. Submissions are divided into host submissions (already recorded audio or third-party links, host profile, Jianshu articles) and author submissions (author profile and articles. If adopted, Jianshu Podcast will record the audio), and then the audio will be played through the cooperating Himalaya Radio.

To sum up the articles and audios that have been published so far, their styles are basically chicken soup and emotional articles. The most important point is that the audio must be a female voice.

2. Machine Algorithms

In fact, the content control on Jianshu mainly relies on machine algorithms. The content controlled by machine algorithms includes popular articles (including seven-day hot topics, thirty-day hot topics, and new items) and special topics (popular and recommended) under the article section. Next, we will analyze them one by one.

①Popularity: In fact, the algorithms for popularity, seven-day popularity, and thirty-day popularity are the same. Assume that A is the score of a single article, a is the reading time of a single article, b is the total number of readings of a single article, c is the number of comments on a single article, d is the number of likes of a single article, e is the number of rewards for a single article, and n1, n2, n3, n4, and n5 are different weights given to e, a, b, c, and d respectively. We can list a formula: A=n1*a+n2*b+n3*c+n4*d+n4*e. The higher the A score, the higher the ranking of this article. The only difference between these four is that the time span of their calculation is different. Popularity should calculate the score within one day, while seven-day popularity and thirty-day popularity should calculate the scores within seven days and thirty days.

At present, the calculation method for new entries on the list has yet to be determined. From what I have observed, the list is refreshed every 10-20 minutes, sometimes once an hour, and it may be calculated based on a shorter time span.

②Topics: The topic section is divided into two sub-sections: Popular and Recommended. Popular and recommended are displayed in the same way, showing topics to users, and users can choose to follow them. There is also an algorithm for ranking popular posts: it is estimated that a score is obtained based on the number of posts published and the number of followers obtained, and it is still unknown how the weight of the two is calculated. The algorithm for ranking recommendations is estimated to be a score calculated by the number of followers + the number of followers + the articles (topics) you read. The specific calculation method is still unknown. However, both may involve human involvement.

In addition, I would like to raise a separate issue here: topic content (articles under the topic). Before May, topics were created by individuals, who submitted content to the creators, who were responsible for reviewing the content. In this way, the standard of content was completely determined by the individuals, and it was impossible to make a unified answer. However, in June, Jianshu officially launched a plan to recruit topic editors, preparing to standardize topic content. Therefore, at present, the review of topic content (articles under the topic) can be said to have official involvement, which can be summarized as the scope of manual review mentioned above, but it is not known how it is implemented at present. Therefore it is listed separately here at the end. According to current official actions, Jianshu's special topics are a direction that we are currently focusing on, and should be aimed at building a community.

4. Push and recommendation strategies

Push. Here, I only analyze the push of the product itself. The push methods of third-party new media such as WeChat , Weibo, and email are not within the scope of my analysis. Jianshu’s push and recommendations are divided into two categories: daily reports and special topics.

I don't want to explain too much about the concepts of recommendation and push here. If it is push, it is definitely recommended to users, but if it is recommended content, it can be delivered to users in different ways. This is the relationship between the two. Recommendation focuses on the standards for content selection, while push focuses on grasping the rules of push. Therefore, if we talk about these two separately, it may seem like we are too obsessed with concepts and neglect analyzing the facts. Therefore, the two methods mentioned below both include push rules and standards for recommended content.

1. Daily News

As mentioned before, the daily newspaper is displayed as a separate presentation under the article section. There are two types: Jianshu Morning News and Jianshu Evening News. The push time is approximately 11:00 to 12:00 and 20:00 to 21:00 in the evening, with a floating time of about 30 minutes before and after. Due to the function of the mobile phone, the push page cannot be captured, so the specific push picture cannot be posted.

Through summarizing and inducing, it can be found that the recommended content of Jianshu Daily follows a regular pattern. It is divided into morning paper and daily paper, each containing 8 articles. The 8 articles are arranged from top to bottom (the order is not yet known). Each article reveals the title , author and some article information. Users who are interested can click to enter the article to read the full text. The types of articles include: education, two-dimensional, film reviews, music, history, novels, philosophy, characters, lifestyle, food, serials, finance , art reviews, fitness, games , worldly affairs, etc., and there is no repetition every day (it feels like there is everything). If the amount of statistical data is large enough, I believe there must be a pattern to follow (this can reflect the value of operations). However, due to time constraints (late stage of laziness), due to lack of information, I can only analyze it to this point.

There are two sources of recommended content: one is official selection. The second is recommendation by the special editor. The first one is easier to understand, which is that Jianshu officials screen the submitted articles and then push them. The second is that users who have applied to become special topic editors can submit articles to Jianshu Daily as special topic editors. As for the selection criteria, the type of article mentioned above is one of the criteria, and the rest of the criteria have been mentioned before and will not be repeated here (manual review section).

2. Pay attention to push

Follow push means that the user chooses to follow an author or topic, and then the user can choose whether to receive the updated article push as soon as possible. If you choose to receive, then after the author or topic you follow updates an article, the user will receive a push notification of the updated article.

So far, the above is the analysis of the operating mechanism of Jianshu content. In summary, we can draw the following conclusion: whether it is hot topics (seven days, thirty days, etc.), special topics, daily reports, or Jianshu podcasts, every presentation method, every control method, and manual review are all carried out for the goal of helping users find the content they want to see and providing authors with opportunities to show themselves. This may be the core competitiveness of Jianshu in the future. When there is a surplus of content, it is valuable to obtain the desired content with less time cost.

Mobile application product promotion service: APP promotion service Qinggua Media advertising

The author of this article @贾楚 is compiled and published by (Qinggua Media). Please indicate the author information and source when reprinting! Site Map