8 data points to gain insights into Bilibili

Recently, a friend @刘昊a on Zhihu used a crawler to do a data analysis on Bilibili, and found some of the findings to be very interesting.

Based on his statistical results, I made some simplifications and added some of my own ideas.

1. Creator submissions

Looking at the number of submissions over the years, there has been a huge increase since 2016. As of mid-March 2020, there have been 11.58 million submissions. Based on this estimate, the total number for the whole year can reach 40 million. Consider that Bilibili’s daily active users are only over 5000w, so the ratio is almost 1:1.

Judging from the introduction of creators, the growth rate is astonishing. From 2017 to 2019, it has been growing at a rate of doubling annually.

Another perspective is to look at the retention of creators. The retention rate in the chart is calculated based on creators who continued to submit in 2020 after their first submission in year X; and creators who submitted more than 15 pieces of work each year after their first submission in year X and continued to submit in 2020.

This number is also encouraging. About half of the people who have submitted manuscripts are still submitting.

2. Submissions by Region

I have been curious about the situation in each section before, and I wonder which of the major sections, games, life, and entertainment, has the most submissions. According to the submission statistics from 2018 to 2019, it is as follows:

Judging from the number of submissions, the gaming area is still far ahead, and the two-dimensional attribute is still the majority. However, the life area has experienced an explosive growth of almost two times, which also indirectly illustrates the direction of B station's breaking out of the circle: popularization and ordinary people. If most people want to upload some videos, the living area is probably the most suitable place.

When it comes to the lifestyle area, different types of videos have different ratios of submissions and plays. For example, comedy and food videos have high leverage (the average number of plays per submission is higher).

The game area is like this. GMV and Mugen may be highly professional content, and the head phenomenon is very obvious. (GMV is Game MV, which is a MV based on games or CG; Mugen is an engine for 2D fighting open platforms, and here it refers to related videos)

The leverage in the science and technology zone is also very obvious. Excellent technology video with a high average number of views. This is another main battlefield for breaking the circle besides the living area, and the content of this battlefield is of high value, and it is possible that it will be a long-term seller (rather than a best-seller).

3. Header Content Partition

The proportion of the partitions where videos with over 1 million views are located is as follows:

As expected, the lifestyle area ranks first, followed closely by old games and animation series.

The second-level categories are as follows:

Judging from this classification, two-dimensional and non-two-dimensional content are no longer distinguished and are mixed together.

4. Top up main partition (ranked by number of fans)

Looking at the pie chart, the top up masters’ partitions seem to have not changed from 2018 to 2020. However, if we look at the changes in proportion in a different way, we can see the trend:

The proportion of living areas and technological areas has increased dramatically. These two are also our daily impressions of the so-called main battlefield of B station breaking out of the circle. The rest of the increases are mostly unrelated to the two-dimensional world.

5. Head up main flow

Looking at the percentage of the top X up masters' contribution to the platform, the results are as follows:

This is a perfect illustration of the 2/8 rule . Pay attention to the 6th column, which shows the top 113,000 up-hosts. They contribute 20% of the total submissions on the site, but account for 75.62% of the total playback volume.

What is the liquidity of these up masters? Will there be a problem of class stratification? The statistical results of the changes are:

Surprisingly, this change ratio is very close to the figures of the United States, France and other countries used in The Great Gatsby Curve when discussing the issue of social class mobility, that is, between 40% and 50%.

Of course, there is not much comparability here. When discussing social class mobility, the time span is intergenerational, which means that children are influenced by their parents. The above statistics on B station up masters are based on years, and the pace is much faster. A lot of changes may happen in a few years.

From my personal observation, the top mobility of Bilibili is still very high, and it is far from the point where the top up-masters monopolize the content. Opportunities for newcomers not only exist, but are huge.

6. User Likes and Ratings

The ratio of approval to evaluation reflects the user activity to a certain extent. Liking is a light interaction, while commenting is a heavy interaction; liking merely expresses approval, while commenting involves participating in discussions and expressing more content.

From the chart, we can see that the more TOP the video is, the higher the likes and comments ratio is. In other words, the more popular the video is, the more light interactions it has.

What’s interesting is that the like-to-rating ratio for the entire site is 3.46, which is much lower than that of Douyin and Kuaishou (Kuaishou is lower than Douyin, which is also consistent with the observation result). This also indirectly confirms our daily feeling: Bilibili users are more active and tend to focus on interaction.

The original author also mentioned that Bilibili also has other interactive methods such as coin insertion and collection. If all are taken into account, the activity level will be higher.

7. Video Length

The average length of videos on the entire site was around 20 minutes in the early years, and has been declining significantly since 2018. In 2020, it’s already 12 minutes.

Judging from the current video length distribution, videos of 1min-3min account for the largest proportion, followed by 3min-5min, and then 5min-10min.

8. Video Lifecycle

This is the distribution of comments over the time dimension. The horizontal axis time represents the difference between the comment time and the video release time. 0 corresponds to 49.54%, which means that 49.54% of the comments were made on the day the video was released. More than two years ago corresponds to 2.56%, which means that 2.56% of the comments were made more than two years ago.

As can be seen from the chart, the video life cycle of Bilibili is very long. Although there will be an immediate explosion when it is first released, in the long run, there will continue to be users watching and interacting. Of course, since there is no comparison with content platforms such as Douyin and Kuaishou, it is difficult to say whether this number is high or low.

I am personally very interested in another related data, but the original author did not count it: Are the video life cycles in different partitions the same? For example, whether the videos in the technology section can be “long-selling”. This data statistic must be of great value.

The above are 8 key data and conclusions that I have compiled that are of personal interest. I hope this can inspire you.

Author: Liu Yanfei

Source: Liu Yanfei Notes