The Future Depends On The Data: How Does Clubhouse Build TikTok-Style Algorithms?
The future fate of Clubhouse depends on how it helps users find great content and increase participation. Although facing this problem, everyone can think of the correct answer-“user data” but the specific implementation is definitely not so easy.
For people like me who have worked for Snap, LinkedIn and founded a social media analysis company, the potential of Clubhouse can be said to be beyond doubt. With the arrival of celebrities such as Oprah, Zuckerberg, Musk and even Drake, this “embedded audio chat” application has officially entered the mainstream.
However, the future fate of Clubhouse depends on how it helps users find great content and increase participation. Although facing this problem, everyone can think of the correct answer-“user data”, but the specific implementation is definitely not so easy.
Table of Contents
“What is Clubhouse?”
Let’s listen to what users say about Clubhouse:
- Not suitable for elderly users
- The next Snapchat in audio
- Zoom with a community function
- The “radio” of the younger generation
- The first AirPods community network
- The equivalent of random chatting on the chat hotline in the 1990s
- All-inclusive social interaction layer
Essentially, the function of Clubhouse is to guide users to join the audio room and chat with other people. Some chat rooms are places for moderators to chat with private friends, and some rooms may have thousands of users listening to Oprah’s discussions. The Clubhouse has achieved amazing user growth in just a few months, and its current valuation has reached 1 billion US dollars.
Many people attribute this success to our urgent need to restore social interaction during the social quarantine triggered by the epidemic. Live broadcasts and well-produced videos on Twitch or YouTube are obviously difficult to meet this demand. Having said that, this exponential growth does not seem to stem from this alone.
Clubhouse lowered the barriers to participation in the dialogue. Writing blog posts and creating beautiful videos often requires a lot of energy and technical investment. In contrast, participating in a voice chat is even easier than taking a selfie. Setting up a one-time chat room is also easier than planning a podcast program. All in all, this app greatly simplifies the creation of talk broadcast content.
Clubhouse, The Internet Version 1.0 Of The Radio Hotline
Ten years before the birth of the Internet browser, in the 1980s, radio hotlines were gaining momentum. Its popularity is very similar to that of Clubhouse: it lowers the threshold for content delivery and dissemination.
The AM frequency band provides more abundant bandwidth to publish all kinds of content. In addition, lose control policies have also allowed local radio stations to develop into industry-scale content creation platforms rapidly.
Forty years later, the popularity of the Internet has brought us a similar effect: publishing content at zero marginal cost; and taking Clubhouse as an example, there are no specific broadcast or discussion topics restrictions. @benthompson emphasized in social network 2.0 that Clubhouse has typical Internet 1.0 features:
Please note that the essential feature of v1 digital products is that they copy the content that already exists offline. For Facebook, this means digitizing the connection between friends and family; for Twitter, it is equivalent to digitizing the connection between friends and family. Easy chat is moved to a public platform. The reason all this can only be called v1 is that in the v2 product, certain functions that can only be achieved through the unique attribute of digitization must be included.”
Conversely, it can be seen that Clubhouse did not make full use of this “digital property”. Fundamentally speaking, all social applications need to solve two problems: 1) find social interaction; 2) start social interaction. As long as you walk around the Clubhouse for a while, everyone will realize that it is difficult for you to find interesting conversations here. Of course, you can follow the users who mainly chat with certain content, but the primary method is still different from the recommendation scheme that Twitter provides exciting options.
We can understand Clubhouse as a social graph structure, which is Twitter’s most clumsy information management method. A friend recently told me: “I finally set up a Twitter account, and now I can access only the content I am interested in.”
He has been using Twitter since 2008 and only now has he really found it in line with his preferences. Usage! Twitter will often ask users to update and adjust their followers and “customer satisfaction” options. What’s more frightening is that we tend to see a lot of content on Twitter that we don’t want to watch at all; similarly, users often get stuck in long and tedious chats on Clubhouse.
The Problem Is The Challenge Of Data.
In order to find the most suitable topic for the current user, Clubhouse needs to understand all the content of the conversation, that is, to transcribe the content of the conversation in real-time. This goal is technically feasible. The current speech-to-text algorithm is already very powerful and does not have a high threshold for use.
Clubhouse currently does not record the content of the user’s conversation; I think this is because the operator temporarily does not want to affect the authenticity of the conversation. Similarly, Snap also quickly became popular at first. Please note, however, that Snap and Clubhouse have never made it clear that they will not keep records of information.
TikTok’s Recommended Magic
As today’s number one algorithm manufacturer, TikTok has shown a way out for Clubhouse’s subsequent development with its powerful recommendation function. Unlike Clubhouse or Facebook, TikTok is not primarily dependent on social graphs but a powerful recommendation engine. As @eugenewei mentioned in his talk about TikTok and the “sorting hat”:
“Bytedance software engineers have paranoid attention to their algorithms and use half of the inspection cycle in this regard. Bytedance is also known as an algorithm company, and it was the first breakthrough “news” application, apps. Today’s Toutiao, followed by the short video platform Douyin, and finally TikTok. ”
TikTok’s machine learning magic comes from a supervised learning system that can quickly find videos that users are interested in. It measures how long users have watched, how fast they scroll and considers various other behavioural characteristics to improve the accuracy of the recommendation algorithm as much as possible.
In addition, the algorithm evaluates the actual attributes of the video clip: content, speed, sound, colour, text, etc. In order for the algorithm to be effective, TikTok needs a large amount of data. To create a system that can recognize image content alone, we need 14 million classification training images.
Whether to collect data has always been an important choice at the product design level. When designing the TikTok core, byte beating took this aspect into consideration. Today, the algorithm is far superior to Facebook’s recommendation algorithm. In addition, this part of the data has also established huge business barriers; even Facebook is difficult to easily overcome.
TikTok is an algorithm-first company. It does not rely on social graphs and independently builds a powerful recommendation engine.
With the development of Clubhouse, operators also need to make design choices in this regard. How should they record the positive emotions in the conversation? The Clubhouse is a passive carrier, so users may not directly respond to conversations.
The problems faced by vendors such as Spotify or Deezer are relatively simple. For example, they can easily find that users do not like “cutting songs”, and other operations are delayed, but Clubhouse can’t, and it’s hard for them to know whether users in the chat room like it or not. The current dialogue, which greatly increases the threshold for algorithm improvement.
Recommending content creators may be a clear way to measure user satisfaction, but this method is often slow to work in the early stages of use. According to my own entrepreneurial experience, tracking the spread of the same content on different platforms or the number of reposts is often an ideal signal to measure participation.
This approach seems to also apply to Clubhouse. Today, you can already listen to the exciting passages in the Clubhouse conversation on Twitter or Facebook and other platforms.
Therefore, in the future, Twitter and Facebook are likely to develop products that compete directly with Clubhouse. This is because they have social graphs and control communication channels that can measure the quality of content.
The Future Of Clubhouse
Despite a series of challenges, Clubhouse’s rapid early development momentum is still worthy
of recognition. Whether or not it can persist in such a good state in the future depends on whether Clubhouse can promote the diversification of the published content at a lower cost. The operator is, of course, aware of this problem and therefore announced that it would invest a huge amount of money to provide revenue for authoring tool developers and content creators.
In order to expand the scale of their business, they also need to establish various data and participation indicators. If it fails to break the circle, Clubhouse can only become a set of sub-stacks under the audio content-although it can still be regarded as a success, it will not be able to achieve its full potential.