Internet Community Product Methodology

In the past few years, I have been working in the Internet community. I thought I had done a lot and accumulated some experience. Naturally, I "believed a lot of things, didn't believe a lot of things, and thought that what I knew was everything."

These "beliefs" may not be special, they just make sense to me because I don't know more. Even so, I still have the urge to record these thoughts from time to time.

Recently, by various coincidences, I have had a lot of exchanges with friends who are working on community products or want to work on community products, as well as colleagues who are not sure whether to work on communities. I found that the questions that everyone thinks about and the confusions they have about Internet communities are almost the same.

After once again reiterating some of the same old points, I decided to write this article. This article records my personal thoughts and understandings about the Internet community over the past few years. Although the content has been concise and summarized as much as possible, it is still a bit complicated.

The themes of the articles are similar to "Building an Internet Community from 0 to 1", "X-word long article: mid-game thinking on the Internet community", "Why most companies fail to build Internet communities", but if one of them is used as the title, it always feels a little bit lacking.

A few days ago, I used WeChat Reading and came across "Yu Jun's Product Methodology" which I read at the beginning of the year. I thought that the framework of this article is similar to it, and the content is also subtly influenced by it. Finally, I closed my eyes and decided to use "Product Methodology" as the title to salute Professor Yu Jun.

1. What is an Internet community?

1. Definition of Internet Community

The Chinese word “community” is translated from the English word “Community”. “Community” comes from the Latin word “Cīvitās”, which means group or shared ownership. In the sociological sense, community is generally interpreted as a social group and its activity area that lives in the same geographical area and has a common consciousness and common interests.

In the context of the Internet, the characteristic of the community being "within the same geographical area" has been weakened, and the feature of "having common consciousness and common interests" has evolved into "having the same interests, cultural preferences and values" and has become more prominent.

Therefore, this article summarizes the definition of Internet community as: an online communication space that brings together social groups with similar interests, cultural preferences and values.

(The word "community" was translated by Fei Xiaotong, the author of "Rural China", and is still used today. "Rural China" is also a must-read book for community product managers)

2. Classification of Internet communities

An Internet community is like a basket that can hold anything. If it contains games, it is a game community; if it contains short videos, it is a short video community; if it contains “the story you just made up”, it is... Therefore, there is no single exclusive method to classify communities.

For example: from the perspective of content form, communities can be divided into picture communities (Instagram), video communities (Bilibili), live broadcast communities (Kuaishou), etc.; according to product form, they can be divided into forums (Mop), Weibo (Fanfou), Q&A (Zhihu), etc.; according to different coverage areas, communities can also be divided into vertical communities and comprehensive communities.

A vertical community refers to a community that is oriented towards a specific field or aggregates specific topics, such as the football community Live Bar and the movie community Time.net. Comprehensive communities are not targeted at specific fields or topics. All community products that can be called "national" can be included in the category of comprehensive communities, such as Weibo and Douyin.

Under the same category, the boundaries between different communities are not strictly divided, but are intertwined and each has its own advantages. For example, Instagram was positioned as a photo sharing community in its early days, but later it also added short video content such as Story.

(From left: Bilibili, Weibo, Zhihu)

This article mainly analyzes the community from the classification perspective of vertical communities and comprehensive communities. The reason for this choice is mainly to facilitate the discussion of the two issues that are known as the "annual keywords of the Internet community", namely customer acquisition and growth, breaking the circle and transformation.

1) Customer acquisition and growth

From its birth, germination, growth, to maturity, a community will always face the problems of customer acquisition and growth.

Community designers should first clarify the vertical category to which the community belongs, and then calculate the capacity of the community from an industry perspective, that is, the ceiling of community customer acquisition and growth. Only then can they position the community, design the architecture, and break down and implement the tasks.

Taking the stock vertical community as an example, data from China Securities Depository and Clearing Corporation shows that as of November 2020, the total number of investment accounts in China was 176 million, which is approximately equal to the number of securities investors in China. Can this number be understood as the ceiling of the number of stock community users?

The answer is no. Although there are 176 million existing investment accounts, they include many zombie accounts that no longer trade for various reasons (essentially one reason - loss), and what we need to be clear about is the number of active investors.

Although China Securities Depository and Clearing Corporation has stopped updating the data of active accounts since February 2017.

However, according to past statistical data, the proportion of active accounts to total accounts has basically remained in the range of (10%, 40%). The stock market situation in 2020 is a matter of opinion. If the median proportion of active accounts is taken as 25% and the number of active accounts is calculated, the value is 44.03 million, which is the ceiling of the number of stock community users at this stage.

The emerging stock community can set more reasonable short-, medium- and long-term growth goals based on this figure.

For mature stock communities, if the number of active users in the community itself is close to the industry ceiling, then no matter how hard they try, it will be difficult to make a big breakthrough within the original framework. At this time, community designers naturally have to think about breaking the circle and transformation issues.

(Total number of investment accounts and active accounts in China. Data source: China Securities Depository and Clearing Co., Ltd. website; data range: May 8, 2015 - November 30, 2020)

2) Breaking the circle and transformation

For the vast majority of surviving vertical communities, the ceiling is not out of reach. If the goal is continuous growth, sooner or later they will face the problem of breaking the circle. Breaking the circle/going out of the circle is a term used in fan circles, meaning that a star or a work breaks through the fixed fan circle and becomes known to a wider range.

For example, Wu Tiao Ren, a niche band from Haifeng, Guangdong, whose songs are mostly sung in the Haifeng dialect, broke out of the circle because they were eliminated and rescued repeatedly in variety show competitions.

(Five People)

The breaking of the circle of vertical communities is essentially based on the vertical field and developing in a comprehensive direction related to it. In recent times, the term "breaking the circle" is most often associated with Bilibili.

In the early days, Bilibili was a pure ACG (animation, comics, games) community, featuring barrage and ghost culture, and its core users were two-dimensional enthusiasts; now Bilibili has broken the circle - Bilibili New Year’s Eve Gala, Houlang Video, self-made variety shows and dramas ("Rap New Generation", "The Sky of the Wind Dog Boy") and other content, by the way, Bilibili has broken the original two-dimensional circle and is known to the public.

But long before these official self-made contents went viral, Bilibili had already begun to change its course.

In July 2018, Bilibili COO Li Ni said: "Bilibili currently has more than 7,000 vertical interest circles. The traditional two-dimensional content currently accounts for 30% of the total visits to Bilibili, while diversified interest circles such as life, entertainment, and fashion have become the main content of Bilibili."

Judging from the results, Bilibili has completed the transformation from being based on the two-dimensional world to being based on the "new wave". Its target group is no longer limited to two-dimensional users, but a wider range of young groups. In Q3 2020, Bilibili's monthly active users reached 197 million, a year-on-year increase of 54%.

(From left: Bilibili New Year's Eve Party, Houlang Video, and Rap New Generation)

For vertical communities that are in decline or in sunset industries, the transformation issue is even more serious and may even be a matter of life and death.

In terms of direction, there are only two paths for community transformation in a recession. One is to expand into related fields, and the other is to switch to a new track that can be empowered by existing resources. When it comes to practical aspects, specific analysis is required for specific issues.

3. Community and Social

Haruki Murakami has a book called "What I Talk About When I Talk About Running". The sentence structure of the title imitates his idol Raymond Carver's "What We Talk About When We Talk About Love". This sentence structure is also applicable to the topic we are going to discuss next - what are we talking about when we talk about community and social interaction?

(From left: Haruki Murakami, Carver)

Community and social networking are a pair of easily confused concepts that are often mixed up. The community is an interaction driven by content, which solves the problem of connecting people with content, while social networking is an interaction driven by relationship chains, which solves the problem of connecting people with each other.

In terms of interaction forms, the community's interaction forms include likes/comments/forwards, etc., while social interaction is IM, instant messaging.

For example, typical community products, in addition to those mentioned above, include Douban Groups, Jike, Tieba, etc.; in terms of social products, there is WeChat, which is almost synonymous with social networking in the mobile Internet era, as well as QQ, Momo, Tantan and a number of other products that are named with two-character repetitions as if they had been discussed.

By comparing and analyzing the existing products on the market, we can find that the vast majority of community products and social products generally have both social and community elements. For example, Douban groups, in addition to posting and interacting (community), can also send private messages (social); and WeChat, can both have conversations (social) and browse and post to Moments (community).

Since the product contains both community elements and social elements, how do we distinguish whether a product is a community or a social product?

The answer is to look at the first-level page of the product (which is actually also the focus of the product). The first-level page of Douban group is the topic list, and the private message function is in the second-level page message, so it is a community product; the first-level page of WeChat is the conversation list, and the entrance to the circle of friends is on the second-level page of the discovery tab, so it is a social product.

(From left: Douban group topic list, WeChat conversation list)

2. How to build a community?

1. Community positioning

When we analyze things, we like to trace back to their origins. Even the universe was proven to have originated from a singularity in 1970.

So, what is the singularity of community?

I think it’s positioning. Community product managers encounter many problems in their daily work, such as the community has complete functions but fails to attract users? Is there no difference between the community and competing products? Is the community becoming more and more niche? All can be traced back to the positioning of the community.

(Hawking and Penrose proved that under reasonable conditions, singularity is inevitable)

Community positioning also needs to pay attention to the basic law. Although there are a variety of product positioning methods circulating on the market, the basic ideas basically come from the book "Positioning: The Most Influential Concept in American Marketing Ever" co-authored by Al Ries and Jack Trout.

The book describes the specific steps of positioning work. Simply discussing the method may be too abstract, so we set up a scenario and analyze it with specific cases. The scenario is set as follows: If you were to create a stock community in 2011, how would you position it?

First, analyze the entire external environment and determine "who our competitors are and what our competitors' values ​​are." In 2011, after experiencing a round of bull and bear markets, the A-share market was generally in a unilateral downward trend. The market was sluggish and the activity level dropped significantly compared to 2007 and 2008.

At this time, the communities that investors frequently browsed included Guba and Taoguba. Although the names of the two products differed by only one letter, they were completely different types of community products.

Guba is a stock community under Eastmoney. Its slogan is: A stock-themed community under Eastmoney. The stock bar is aimed at the broadest group of stock investors and does not impose too many restrictions on users. As long as they comply with regulations, users can speak freely in the stock bar.

Taoguba is a "professional A-share investment and stock trading forum exchange and sharing community". Taoguba holds several real-time trading competitions every year, gathering the most active short-term players in the market. Brother Zhao, who is now regarded as a god by investors for his "10,000 times return in 8 years", became famous on Taoguba.

For ordinary users, the purpose of using Taoguba, or the value of Taoguba to users, is to follow the operations of experts and explore the latest market themes.

Secondly, avoid competitors’ strengths in customers’ minds, or take advantage of the weaknesses hidden in their strengths to establish the brand’s advantageous position – positioning.

The stock bar is backed by Eastmoney.com, and has cultivated a user mindset where users can not only speak freely in the stock bar, but also browse a wealth of UGC (User Generated Content), and learn about comprehensive and timely information, announcements, research reports, and other OGC (Occupationally-generated Content).

The shortcomings of the stock bar are also quite obvious:

On the one hand, the relaxed and open communication atmosphere has led to the rampant "argumentative" users in the community. Even relatively professional investment-related discussions may be countered by argumentative users with various words, resulting in a decreasing willingness of high-quality content producers to create, and a poor community discussion atmosphere;

On the other hand, even if the producers of high-quality content have a big heart, ignore the arguments, or respond tirelessly, and continuously output high-quality content, the product structure of the stock bar (strong topic (the topic here refers to stocks) flow + weak attention flow) also determines that the high-quality content cannot get sufficient exposure, and it is difficult to form a professional content ecosystem.

(Stock Bar)

The model of Taoguba's real-time trading competition has determined that there are many short-term experts in the community, and their exploration of topics and understanding of logic are ahead of the market. Users can follow the operations of the experts in Taoguba without thinking and gain excess returns, or learn the real-time trading ideas of the experts and put them to use for themselves.

The weaknesses of Taoguba are also obvious. On the one hand, Taoguba relies too much on real-time trading competitions, which results in attracting mostly ultra-short-term users and a narrow user circle. On the other hand, the vast majority of users attracted by Taoguba do not have the ability of top players, and the result of blindly following operations is a shortening of their investment life cycle, leading to a rapid user churn rate on the platform.

(Taoguba)

Therefore, in 2011, if you want to establish a new stock community, you need to avoid the advantages of competitors such as OGC, real-time trading competitions, etc. When choosing the target customer group, you need to avoid customer positioning that is too narrow or too broad. The rest are all viable paths that may establish a unique customer mindset.

For example: establish a PGC (Professional Generated Content) community, or a portfolio-centered community, or a community for value investors, etc.

Combining actual cases, Snowball, a stock community launched in November 2011, positioned itself as “smart investors are here”, cleverly avoiding the advantages of Stock Bar and Taoguba, and choosing the direction where it can establish its own advantages - "it (Snowball) does not recommend stocks, but relies on users to produce content, so that investors who like independent thinking can gather here to study companies and talk about investment together" (Fang Sanwen, CEO of Snowball).

(Snowball)

Again, seek a reliable proof for this positioning - a letter of trust. After the positioning is determined, Snowball mainly builds trust from two dimensions.

First of all, in the stock dimension, Snowball sets the default signature of the individual stock details page to discussion, so that users can understand the latest, hottest and most essential discussions about the stock in real time while browsing the market. Users can discover experts from the perspective of stocks.

Secondly, in terms of people, Snowball has specially designed users' personal homepages. When visiting other people's homepages, you can quickly understand the user's real/simulated portfolio returns, stocks that he is good at discussing, and past speeches.

In addition, users can also dynamically select the people they follow, and through continuous communication with the people they follow, they can know who is good at answering which questions, so as to find people who are better and "smarter" than themselves.

(From left: individual stock details page, user personal homepage, homepage follow list)

Finally, integrate this positioning into all aspects of the company's internal operations, especially have sufficient resources in communication to implant this positioning in the minds of customers.

In terms of internal operations, Snowball adopted a closed invitation registration system in its early days. The operation team first looked for "smart investor" users who matched Snowball's temperament, and then asked these users to invite people with similar temperaments to them, to ensure that the original tone of the Snowball community would not be diluted as it grew bigger and bigger.

In addition, Snowball has set up a special user mining team. Team members will screen high-quality content from a large number of posts, and then actively contact content producers to guide and inspire them to produce more articles about company research, and attract more traffic to the articles through forwarding and other forms.

In terms of external promotion, in addition to inviting high-quality authors for interviews and participating in offline Snowball Carnival events, Snowball will also reprint high-quality content produced by platform users on external platforms such as Weibo and WeChat public accounts.

For articles that have certain topic attributes, Snowball will also include them in Snowball Weekly and publish them on platforms such as WeChat Reading and Kindle. For the systematic content created by high-quality content producers on the platform, Snowball will also proactively help authors contact publishing houses for packaging and publishing.

Through the above work, Snowball not only helps content producers increase their number of fans and influence and promote a positive cycle in the content ecosystem, but also instills the idea in potential users that "smart investors are all on Snowball."

(Snowball creator rights)

2. Product Architecture

After clarifying the product positioning, the next step is to build the product architecture.

There are many nouns similar to "product architecture": "business architecture", "strategic architecture", "information architecture"... These words are like twin brothers, often appearing together, and sometimes being misused, giving people a confusing feeling. In fact, there is some overlap between these concepts, but they are also different.

• Strategic architecture/business architecture: the top-level architecture, including the business operation mechanism including business logic;
• Product architecture: The architecture of product functions and systems that connect strategy and information. Its specific function is to classify and integrate different functions around the goal.
• Information architecture: the most front-end architecture, including basic interaction design, information and other presentation layer architectures.

The three structures are progressive and interconnected. The intersection of strategic architecture and product architecture lies in the organization of business processes, and the intersection of product architecture and information architecture lies in the design of product systems and interface presentation.

This section mainly discusses product architecture. Content related to information architecture will be integrated into subsequent content production, content consumption, etc.

At present, there are not many books and materials on product architecture design on the market. When this happens, some people often choose to use the "panacea" in the field of product managers - the five elements of user experience to apply product architecture design, but the results may be unsatisfactory.

Although Jesse James Garret's "The Elements of User Experience" is a must-read book for product managers, it provides a sufficiently abstract and easy-to-understand product design approach, which is to design products from five levels: strategy, scope, structure, framework, and presentation. However, the five elements of user experience explain the product more from a macro level.

In terms of actual implementation, this idea does not have much guiding significance.

However, by sorting out limited information, we can still abstract a complete product architecture design method.

First, enumerate all the functions of the product and classify and aggregate them from the perspective of goals or meeting certain user needs. For example, the "Me" channel of the interest community aggregates the user's own dynamics, photo albums, archives, footprints and other functions based on the perspective of user-related information and production content.

(Immediately - "Me" channel)

Secondly, sort out the upstream and downstream relationships of functions and form a process.

Communities generally do not involve overly complex upstream and downstream process relationships. From my own experience, the most complex process in a community may be the review process.

As for the frequent feedback from users that they cannot find the blacklist function, it is not because the process to reach the function is too long or the process is confusing. Most of the time, it is just because product managers are worried that if it is too obvious, users will actively try it, so they deliberately design it this way.

If the path is clear, it basically only takes 3 to 4 clicks from the first-level page to the blacklist function.

(Instantly - hidden blacklist function)

Finally, analyze the relationship between different categories of functions or different processes, and design a mechanism for mutual coordination. For example, when designing content production functions, it is necessary to also consider the design of functions related to content consumption.

Taking Moments as an example, when designing the posting function, in addition to considering whether it is possible to add pictures, upload videos and other editing functions, we must also consider how the content is distributed after it is successfully published, on which pages users can browse the published content, and how the content is displayed, whether users can like, comment, and forward it, etc.

(From left: Immediately - content editing page, Immediately - content consumption list)

There are a thousand Hamlets in the eyes of a thousand people. Even if we follow the same set of rules mentioned above to design product architecture for the same proposition, different people may give very different results. Is there a standard to judge the design results of product architecture?

Generally speaking, a product architecture that has the following three characteristics: simple and efficient, easy to understand and use, and highly scalable, can be considered a good product architecture. We will not elaborate on this here. After talking about product positioning and product architecture, let’s talk about the hot topics in the Internet community from the perspective of content.

3. Content Production

From the perspective of producers, content can be divided into three categories: UGC (User Generated Content), PGC (Professional Generated Content) and OGC (Occupationally-generated Content).

For example, I, a newbie, uploaded a video of eating noodles with the lights off on Bilibili, which is UGC; the well-known up-loaders on Bilibili, Huanong Brothers, contributed a video of eating roasted duck at Brothers’ house, which is PGC; the New Year’s Eve party uploaded by the official account of Bilibili is OGC.

In the community context, what we usually talk about mainly refers to UGC, but the content of most communities actually also includes PGC and OGC (the only communities that can be thought of that do not include PGC and OGC are Baidu Tieba and Douban Groups, which are completely decentralized and allow users to create their own bars/groups).

(Relationship between the community and UGC/PGC/OGC)

There are three main sources of OGC: self-production, purchase and crawling. The specific method to choose depends on the community's own positioning, tone and wallet. For the community, OGC is the most stable and controllable, and can provide users with certain expectations, but it has the highest cost.

In addition, OGC is generally distributed in full, which can provide common topics for community users and enhance interaction.

Although both UGC and PGC come from users, for the platform, the former is mainly responsible for increasing activity, while the latter is used to enhance the professionalism of the platform. Although most communities have both UGC and PGC, it is difficult to give equal importance to the two, and they are bound to be top-heavy (the head may be UGC or it may be PGC).

For example, as mentioned in the community positioning section above, the activity level of Stock Bar is significantly higher than other communities, but it has less high-quality content. Although Snowball has more high-quality content, it is difficult to further improve its activity level.

Imagine if Snowball chooses to use force by relaxing its review process or inserting more pan-financial content into the information flow to attract more "not so smart investors", it is likely to lead to the dilution of the platform culture and the emergence of the phenomenon of bad money driving out good money.

This section mainly discusses the production of UGC and PGC and does not involve OGC. In summary, community designers have the following main tasks in content production.

1) Define quality content

Imagine this scenario: you have just downloaded a community app, and after opening it, the interface is blank with only a line of small words in the middle: No content yet, you can try to speak~ At this time, you may either leave directly or try to speak but don’t know what to say. The result may be that you hesitate for a while and send nothing, or you may just type "1" and send it.

Therefore, when building a content production system, the first thing to do is to provide a high-quality content template as a basic reference for user content production. In addition, the definition of high-quality content is actually also the definition of community culture and values. If the positioning is accurate, subsequent work will be more effective.

2) Produce high-quality content

In the early stages of community establishment, who will produce high-quality content will be a problem that must be faced:

One method is for operations or editors to produce content. Although the quality of such content can be guaranteed, or it can be released only when it meets the passing line, the quantity is limited. Another method is to invite qualified high-quality users to produce content. This idea may produce more content, but the production frequency and quality are uncontrollable.

Regardless of which of the above methods is used, or whether both are combined, the main purpose is to provide high-quality content templates for users to refer to. The ultimate goal is to promote the community to form a self-circulating content ecosystem.

After the community content ecosystem is established, the main task of the community officials becomes to continuously provide incremental high-quality topics for users to discuss.

The topic of increments is mentioned here and expanded a little. People often ask: Is XX industry suitable for community building? In fact, it is enough to see whether this industry can continue to provide incremental topics.

For example: Compared with funds, stocks are naturally more suitable for building communities. The listed companies corresponding to the stocks generate various kinds of information every day, including fundamental information, technical information, news information, etc.

On the other hand, a lot of information about funds, such as stocks held and holding ratios, is only updated once a quarter, and only the net value is updated every day. In fact, there is no rain or shine, so what can users talk about?

Therefore, we can see that most stock communities do not emphasize operations, while fund communities are just the opposite, with voting, topics, mini-games and other content updated every day.

(From left: Stock community, Fund community)

3) Mark quality content

Before the community goes online, a complete content labeling system should be established to screen out high-quality content and give it traffic exposure.

There are many ways to mark content. It can be done through manual logic, such as reviewers marking content during the review, or establishing a special operations team to mark high-quality content. It can also be done through system automation logic, such as the machine automatically calculating the popularity based on interactive data, and marking content that exceeds a certain popularity value as high-quality. It can also be done through human-machine matching logic, where the machine first screens out a pool of high-quality content candidates, and then humans select the high-quality content to improve efficiency.

There is no absolute good or bad in the above methods. If you pursue efficiency, you can use all machines; if you pursue precision, you can use all manual labor.

4) Encourage high-quality content

The incentive mechanism can provide content producers with the motivation to continue creating and is an important part of the smooth operation of the content ecosystem. There are two main types of incentives: spiritual and material. In practice, they are divided into the following three categories:

1. First, traffic tilting, by increasing the exposure of high-quality content through operations such as top placement, active recommendation in information flow, and expanding the distribution range;
2. Second, it provides differentiated functions, such as columns, live broadcasts and other permissions, as well as medals, V certification and other certifications, to enhance the recognition of high-quality content producers for the platform;
3. The third is to monetize creation. In addition to basic operations such as rewards and opening advertising permissions, it can also help high-quality content producers create online courses (Zhihu), publish physical books (Snowball), etc.

(From left: Zhihu, Xueqiu)

4. Content review

Content review should be the most important concern for organizations planning to enter the community field. Before talking specifically about review, we need to clarify the conceptual scope of content review.

Generally speaking, audits are divided into two categories: compliance audits and quality audits. The content review we usually talk about mainly refers to compliance review, that is, reviewing whether the content violates national laws and regulations. Quality review is the judgment of content quality, which is related to content operation and will not be expanded in this article.

Compliance audits are the bottom line for the community. Only content that has passed compliance review can be displayed on the front end. Perhaps because it is too costly to maintain a manual review team, or because manual review may result in errors or omissions, many people’s questions about content review generally focus on whether machine review can replace manual review?

As far as the current situation is concerned, machines still cannot replace human labor. The reason is very simple. Machines are not as smart as everyone thinks. Just like Siri, a top artificial intelligence robot, if the questions you ask are slightly out of the scope it gives, its "IQ" will drop sharply and it will only tell jokes like a fool.

Based on what we have learned before, most companies involved in UGC currently still use manual methods to review content. What the machines do now is more of the first step of keyword filtering. For example, if certain keywords are mentioned, the content must be in violation of the regulations. The machines will filter out these contents directly without going through the manual step.

For the content review process, the following three steps are necessary: ​​machine review, manual review, and manual quality inspection.

As mentioned above, the main function of machine review is keyword filtering, which can filter out 100% of illegal posts and reduce the workload of manual review; the secondary function is content labeling (for video review, the machine will also randomly cut frames of the video). The dimensions of labels are varied, and the goal is always to improve the efficiency of manual review.

When the content reaches the manual review stage, what the reviewer sees is content that has already been processed by the machine. Auditors can quickly judge the compliance of the content by browsing the original content and the auxiliary judgment information provided by the machine. Manual review is bound to result in errors, so communities generally add quality checks, which involves randomly selecting reviewed content for re-review to ensure the compliance of the content.

5. Content consumption

According to the content distribution logic, the content consumed by users in the community can be divided into three categories: topic information flow, attention information flow, and recommendation information flow.

1) Topic information flow

It is a list of content aggregated by topic, such as a list of posts in a Douban group. The sorting method of topic information flow is relatively simple, which is nothing more than the latest post order, the latest update/comment order, and the popular order.

The first two are relatively easy to understand. The popular order is a sorting method that uses a popularity algorithm to filter out recent popular posts and sort them by popularity value. Some people may be curious about the specific popularity algorithm. In fact, the formula commonly used in the industry is "heat value = (initial heat value + interactive heat value) * time decay factor".

To give a more specific example, the above formula can be refined into: Heat value = (initial heat value + (reading volume * x + forwarding volume * y + sharing volume * z + comment volume * a + like volume * b + collection rate * c + triggered attention volume * d + author's personal homepage opening volume * e))/ (e^(k*(t1-t0))), where "initial heat value" is related to the author's attributes, content attributes and entity attributes, and "(e^(k*(t1-t0)))" is the time decay factor based on Newton's law of cooling.

2) Pay attention to the information flow

It is a content list consisting of information posted by people the user follows. Following is an action actively triggered by the user. In theory, if a user follows all the information in the information flow, he or she will want to read it. Therefore, the early information flows were sorted by the most recent release time.

However, as the number of people a user follows increases, the user's follow list page becomes redundant. Even if users have not clicked on the content of a person they follow for a long time, and rarely take the initiative to unfollow that person, this situation may result in the content of people the users are really interested in not being exposed, and the overall stickiness of the platform is reduced.

As a result, Weibo was the first to change the default sorting of the follow-up information flow from the most recent posting time to an intelligent recommendation order based on multiple dimensions such as users' liking for followers and content quality. Although it was criticized, Weibo still insisted on this change with the spirit of "I will go forward even if there are thousands of people against me."

In 2020, WeChat official accounts also made similar optimizations.

(From left: Weibo information flow, WeChat public account article flow)

3) Recommended information flow

In the early days, the community only had one way to distribute content: topic information flow. The emergence of RSS subscription relationships has led to the emergence of information flow. With the rise of SNS social networks and the arrival of the era of national self-media, the amount of content has experienced explosive growth, and personalized recommendation information flows have emerged.

In recent years, recommended information flow has been a hot topic in the community, but not all communities are suitable for recommended information flow.

First of all, personalized recommendations need to be based on massive amounts of content, and the more personalized the recommended content is, the higher the requirement for the amount of content.

For example: If a community has 10,000 posts a day, according to the 80/20 rule, there are about 2,000 pieces of content that can be recommended. If there are 100 content tags, then on average there are only about 20 pieces of content per tag. Users who are tagged with a certain tag only need to refresh once on the mobile terminal to browse all personalized content.

Secondly, the purpose of the recommended information flow is to guide users to immerse themselves, but the premise is that the recommended content has a low consumption threshold.

If the community content itself is professional and consuming one article takes a certain amount of time and effort, it will be difficult for users to stay in this recommendation flow. For example, TikTok, which is now known as a time black hole, mainly provides entertaining short videos as its main content type. Users can watch videos one after another without using their brains, and time passes by while they are immersed in them.

Again, the recommended information flow provides unknown and uncertain content. In product architecture design, the recommended information flow is often placed on the discovery channel, which is essentially anti-efficient.

As for the slightly mysterious recommendation algorithm, it is actually not complicated. In a word, it is "birds of a feather flock together, and people are divided into groups."

• Birds of a bird of a feather flock: It is to recommend similarity based on content attributes, such as from the author or content level (category, tag, keyword, etc.), the TF-IDF method is used to calculate the similarity between contents, and recommend to users with high similarity content that they have clicked.
• People use group classification: it is to collaboratively filter and recommend based on user behavior. The basic idea is to use user behavior as characteristics, calculate user similarity and item similarity, and match information.

Typical examples are that both user A and user C read content 1 and content 3, and user B read content 2. Based on reading behavior, user A and user C are more similar.

Therefore, after user A has read the content 4, content 4 can be recommended to user C. For more detailed algorithm recommendation content, please refer to Xiang Liang's "Recommended System Practice". It is rumored that ByteDance's most original recommendation algorithm logic comes from this book.

3. Final Thoughts

Finally finished writing. I didn’t plan to write so much at the beginning, but I accidentally made a big statement and didn’t hug me when I wrote it. But even so, this article still has not covered content related to community product operation, such as content operation backend construction and community atmosphere construction. Please find time to write it in the future (in fact, it should not be written again/manual dog head).

Because of this article, I took all the time after get off work, weekends, and New Year's Day holidays in the past month, and I had a serious lack of sleep. The only thing I think about now is that I can finally go to bed early today. May the force be with me.

Author: xuxinplus

Source: xuxinplus