TikTok does everything it can to show relevant videos most suited to a viewers interests. As you use your account to view content you will naturally watch some videos longer, like certain types of videos and follow channels that interest you. This teaches the TikTok machine learning systems what your interests are and what you are likely to appreciate viewing in the future.
When a new video is published the platform will analyse the following factors to classify the video, to be shown to the most appropriate demographic.
- Hash tags – these are specifically used by creators to self-categorise the content. If a viewer likes a video with a certain hashtag other videos with the same tag may also be relevant.
- Music/Audio analysis – certain videos use certain audio tracks which makes it easy to classify them as a particular type of video whether that be a dance shuffle or a comedy sketch.
- Video analysis – TikTok will take frames from the video and run these through a machine learning model to match them to existing videos. If your video contains a cat then the machine learning model will identify that the images in the video bare a relevance match to other cat videos. It is much easier and more efficient to do machine learning analysis on still images at scale so it is likely that keyframes are used rather than the complete video in this process.
- Account relevance – If your account is liked by a lot of people interested in a specific niche then it’s likely that your next video may also be relevant to that niche.
- Location – TikTok distributes more content to viewers in close proximity of the content creator. If you are in South Africa you are going to naturally see more content from South African profiles.