Advertisment

TikTok explains algorithm behind recommendations

author-image
Social Samosa
Updated On
New Update
TikTok recommendations

Follower count and previous high-performing videos are not direct factors in the TikTok system that powers the recommendations #ForYou.

When a user opens TikTok and lands on the For You feed, they are presented with a stream of video curated as per their interests. It makes it easy for them to find content and creators. The section enables connection and discovery, which is central to the experience on the platform. It is powered by a recommendation system that tailors the content on display according to the user's interest. TikTok has recently revealed the algorithm behind these recommendations.

Core of recommendations

The recommendation system takes into account user preferences as expressed through interactions with the app. It could about a comment the user makes, the account they follow or the content they like or prefer to skip. These actions are considered as signals to the system, helping create a personalized experience.

Multiple factors are at play. Broadly, they include user interactions, video information and device and account settings.

User interactions refer to the way the user interacts with the video. Likes, shares, follows, comments and content the user creates are accounted for. Video information includes details like captions, sounds and hashtags. This information is also recorded from the way the user interacts with the Discover Tab.

Device and account settings include language preference, country settings and device type. These details are important from the perspective of optimisation for performance but they have a lower weight in the recommendation system compared to other data points.

If a user finishes watching a longer video from beginning to end, it is considered to be a strong indicator of interest. It is more important as a factor than whether the viewer and the creator are both in the same country. The factors are processed by the recommendation system and weighted based on their value to a user.

Interestingly, while a video is likely to receive more views if posted by an account that has more followers, neither the follower count not whether the account has had previous high-performing videos are direct factors in the recommendation system.

Personalising the For You feed

When a user joins TikTok, they are asked to select categories of interest to help tailor recommendations to their preferences. This helps the system to develop an initial feed, which is gradually polished based on their interactions (likes, comments, replays) on the early set of videos. For users who don't select categories, a generalised feed is created to get started.

When a user comes across a piece of content they don't like, they have the option to long-press on the video and tap on 'Not interested' to indicate the same. They can also hide a particular kind of content or report if they feel it is defaulting the guidelines. These actions contribute the recommendations.

Also Read: TikTok Marketing 101: How to plan a TikTok campaign?

Challenges of recommendations engines

TikTok feels an inherent challenge is that recommendations can limit the experience of a user. It creates a 'filter bubble'. Optimising the feed as per personalised preferences and relevance to the particular individual can risk them to consume a homogenous stream of videos.

The system often attempts to not show two videos in a row that are made using the same sound or created by the same creator. Duplicated content, as well as the kind that can be considered spam, is often not recommended.

Videos that were well received by other users who share similar interests are often recommended. Bringing in diversity in recommendations is something TikTok is trying to work on intentionally. This involves giving users the opportunity to stumble upon new content categories, new creators and experience new perspectives while scrolling through the feed.

The recommendation system is designed with safety as a consideration. Reviewed content found to be depicting things like graphic medical procedures or legal consumption of regulated goods (that would be shocking if surfaced as a recommended video to a general audience that hasn't opted into such content) is not eligible for recommendations.

Videos that have just been uploaded or are under review and spam content such as videos seeking to artificially increase traffic are also ineligible for the For You feed.

Essentially, the system is designed to continuously improve, correct and learn from the engagement of users with the platform to produce recommendations. User feedback and interactions are key to the process.

tiktok news tiktok recommendations tiktok recommendation system tiktok recommedations algorithm tiktok algorithm