India’s recent multilingual social network, Koo, has generated quite a buzz in the past few months. The platform, which closely resembles the microblogging service Twitter, was launched in March 2020 to encourage online discourse in Indian languages. Koo saw an enormous surge in popularity due to the Indian Government’s standoff with Twitter over the latter’s refusal to block accounts the Centre claimed were spreading fake news related to the February 2021 farmers’ protests. Many politicians, celebrities, and government departments tweeted about their shift to Koo and encouraged others to join. As Koo became the talk of the town, we decided to collect data and characterize the platform based on its users, network dynamics, and content.
Koo can be accessed via its website or as a mobile application on iOS and Android devices. It is presently available in languages English, Hindi, Kannada, Tamil, Telugu, Marathi (on Android and iOS), and Bangla (only on iOS). According to the website, support for Gujarati, Malayalam, Oriya, Punjabi, and Assamese is still in the works. Koo users can create a ‘koo’ (post), ‘rekoo’ it (reshare), rekoo it with a comment, or leave a comment below a koo. Koo has a 400 character limit in contrast to Twitter’s limit of 280 characters. Platform arranges users’ feeds according to their chosen language, and they can follow other accounts, like their koos, and mention them. The image below shows how the Koo mobile app appears on iOS.
We collected user profiles, the follower-following network, and content (koos, rekoos, comments, likes, and mentions) from Koo’s public API. We seeded the search with Koo’s official language accounts and a list of popular accounts Koo recommends a user to follow. We then collected the follower and followee IDs of these accounts in a breadth-first manner. We further queried for the user profiles and their content using these user IDs. Our data collection process spanned two weeks, from 26th February 2021 to 11th March 2021. Overall, we collected 4,061,735 user profiles, 163,117,465 follower-following relationships, and 15,375,289 posts (koos, rekoos, rekoos with comments, and comments).
We also created an identity resolution dataset which we can use to compare user behaviour on Koo and Twitter. We collected 38,711 user IDs on the two platforms that correspond to the same entity.
We plot the user joining timeline and observe a massive spike of 700,000+ users, of the 4.07 million in our dataset, in the week of 8th-15th February. 200,000+ users joined on 11th February itself, one day after MeitY tweeted about Koo, which highlights the hype Koo created in a short span of time. We also observed a surge in users around the time Koo won the ‘Aatmanirbhar Bharat App Innovation Challenge’ last year.
Of the 18.1% of the users who specified their gender on their profile, 92.1% identify as male, 7.5% as female and 0.4% as others. We find that females are more active than other genders on the platform in posting rekoos, giving likes, and their follower counts, despite being present in fewer numbers. We find that Koo users are most active in the late evening in terms of posting times, as maximum posts are made around 9 pm IST. The median age of users on the platform is 28 years, which is consistent across all genders.
Users can choose their location from a list of Indian cities; we find that Bengaluru appears most frequently in our dataset, presumably because the platform is headquartered there and was initially available in Kannada. However, Kannada is not the most popular language on the platform, as Hindi and English outnumber it. Tamil, Telugu and Marathi also constitute a considerable portion of the platform’s activity, indicating a degree of success of Koo in promoting Indian languages. Below, we see the language distribution amongst users and posts on the platform.
|Language||Number of Users||Percentage of Users||Number of Posts||Percentage of Posts|
We plot a word cloud of the user bios and observe the presence of words pertaining to state and national identities such as “Tamil”, “Marathi”, “Gujarati”, and “Indian”. This indicates a strong sense of belonging to regional and linguistic identities, in line with Koo’s objective of promoting a local and Indian community.
We analyze the content on the platform based on its hashtags. We see the frequent occurrence of hashtags like “#kooforindia”, “#bantwitter”, and “#koovstwitter”, which project a sentiment of competition between Twitter and Koo, and promote the Koo platform. We also observe hashtags like “#indiawithmodi”, “#atmanirbharbharat”, “#modi”, “#modistrikesback”, and “#bjp” that are associated with the Bhartiya Janata Party (BJP).
We plot word clouds of the top occurring unigrams and bigrams and observe the frequent occurrences of Hindu-centric words like “जयश्रीराम”, “राम-राम”, and “रामपाल-जी”. A majority of the words are in Hindi, indicating its dominance over other languages on the platform.
We find many famous and influential people and organizations on Koo. Many of these are amongst the most frequently mentioned and liked users. “republic”, which is an Indian news channel that has an editorial partnership with Koo, is the most mentioned account. Union ministers Ravi Shankar Prasad and Piyush Goyal are also amongst the most prominent users. Below, we see the lists of most mentioned and most liked users.
On studying Koo’s following network between verified users, we see distinct communities based on language, as users are more likely to interact with others who use the same language as them. English speaking users are more centrally placed in the network, with connections to both Hindi and Kannada speakers. The latter, however, do not have many connections between each other.
Our characterization of Koo produces valuable insights that pave the way for a deeper study of the multilingual platform’s dynamics and its diverse Indian user base. We make our dataset public which can be utilized for future research on Koo and its comparison with Twitter.
Short summary video below and the dataset can be found here.
To read our analysis in detail, you can check out the full report here.
Researchers involved in the analysis — Asmit Singh, Chirag Jain, Jivitesh Jain, Rishi Raj Jain, Shradha Sehgal, Dr. Ponnurangam Kumaraguru