Core Concepts
The TikTok Research API provides a window into the global video trends and engagement patterns on the platform, revealing insights about its user base, content virality, and potential biases.
Abstract
The authors collected and analyzed a random sample of over 500,000 TikTok videos posted between 2019 and 2023 using the TikTok Research API. Their key findings include:
The API failed to meet the expected quotas, delivering at most 65% of the requested videos, likely due to private or deleted content.
The video posting patterns show temporal biases, with uneven distributions across days of the month, days of the week, and hours of the day.
The distributions of engagement metrics like views, likes, shares, and comments follow heavy-tailed patterns typical of social media platforms, with a clear growth trend over the years.
The majority of videos in the sample originate from Asian countries, with India being the largest contributor. The USA is the only Western country in the top 10.
Videos that use "viral" hashtags designed to boost visibility on the "For You" page have significantly more views and likes compared to those without such hashtags, though this effect has decreased in recent years.
The prevalence of conspiracy-related hashtags is low, estimated to be in the order of 0.001% globally and slightly higher in the USA.
The authors highlight the limitations of the API, including the lack of transparency in the internal mechanisms and the inability to access removed content, which constrain the research possibilities. They emphasize the need for improved data access and quality to enable more comprehensive studies of discourse and content moderation on TikTok.
Stats
The number of videos obtained per month was lower than the expected quota, ranging from 500 to 15,000 videos.
The number of distinct users in the collection is almost the same as the number of videos, with only 0.34% repeated users.