Uncovering Chingari’s Feed Recommendation Model


Chingari is on its mission to become the most prominent super entertainment app. It is a unique and fun short video app to create engaging videos, share them, and connect with our global community.

‘For You’ feed is one of the app’s core features and is integral to promoting discovery by surfacing videos based on the user’s interest. Thus, each user will get a unique feed tailored explicitly for that user, making it easier to discover the creators and content they love quickly. This post will explain how your ‘For You’ feed is curated and how you can personalize your experience on Chingari.

Today recommendation systems are a part of our daily online journeys, from e-commerce, where it suggests to buyers articles that could interest them, to online advertising where it surfaces advertisements based on users preferences. For video content platforms like Chingari, it suggests to users the correct content, matching their preferences.

In layman terms, our recommendation system works by considering the interactions data from the in-app interactions such as liking, commenting or sharing a video or following a user and user’s watch history. Based on all these signals, the recommendation system can generate a feed specific to each user, which they will most likely enjoy.

What signals does the recommendation system consider?

Based on the interactions on the app and watch behaviour, the system provides scores to each creator and category to understand the user preference. The signals on which the recommendations are based are-

  • User interactions – likes, shares and comments, the duration watched for each video out of the total duration and the kind of content you create.
  • Creators followed – helps the system surface more videos from similar creators. 
  • Video meta-information includes the video’s description, the video’s category, and the sound used.
  • Language preference helps filter out videos only in the language the user is comfortable in.
  • Location of the device is also considered to promote creators from that location and similar surface videos to new users, which the users from the exact location prefer.
  • The device type is also used to understand preferences from similar device users.

Our recommendation system considers all the above signals to curate a unique feed for each creator. All these signals have different weights that impact the feed, with the highest weights given to user interactions on the app and the lowest to implicit signals like user demographic information. Based on these signals, the system can determine all the videos the user might like and then they are scored based on the engagement to determine the sequence in which the videos will be surfaced to the user.

One exciting challenge which every recommendation system faces is a cold start, which means what to recommend to a new user as there is no past interaction data of the user. It’s integral for a content app like ours to gauge the user’s interest as soon as possible, so our system identifies a cohort of users similar to the new user using implicit signals and surface videos that has performed good for that cohort or most likely achieved good engagement on that cohort. After a few interactions, the recommendation system will consider the watch behaviour and engagement to recommend videos based on the user’s preferences.

The more the user uses the app and watches videos, the recommendation system would receive more information to improve the recommendation to be more relevant to the user.

Another challenge is how to prevent the feed from becoming monotonous as the recommendation system would keep on showing videos only similar to users interests. To prevent this, the recommendation system also surfaces a few videos from other categories that similar users engaged with. The system always balances the ratio of videos that align with user preferences and videos that are not relevant; this is integral for discovery on the platform. 

We have also implemented few logics in the feed to prevent repetitiveness in the feed so that no two videos from the same creator or from the same song should appear consequently or videos that the user has watched before. Also, to safeguard the viewers’ experience, all the videos considered for recommendation undergo a moderation process to prevent content that is Not Safe For Work, such as the use of hate speech or shows use of prohibited substances or scraped spam content from getting surfaced in the For You feed.

To end with, the team at Chingari is working continuously to improve the recommendation models by including more signals and adding models to extract more information about the videos to improve accuracy further. We are continuously looking out for folks to join in our journey towards creating a world class recommendation model, if interested please reach out to us on [email protected].

And this is how your ‘For You’ page works on Chingari – India’s best short video app. Do you have a query? Or would you like to Download the Chingari app and get started with us TODAY! 

Download The Chingari App Now! 


We are looking for the best content creators who can help us in entertaining everyone virtually! And we are continually doing our best to transform Chingari into an entertainment hub. Get started with us if you got a fire in your belly. Download Chingari NOW, and contact us in case of any questions/queries/suggestions at [email protected] along with your Chingari profile ID so that we can check and resolve your issues at the earliest. Also, you can contact us on +91-8591446682/+91-85911 54798.