Using unsupervised AI to analyze tweets from the top Battle Royale games

Getting insights on what gamers are talking about

Fernando Agüero
Building Lang.ai

--

Disclaimer: This article is based on intents discovered by an unsupervised algorithm. There was no human supervision that biased the results. However, the takeaways may be biased for the sake of sharing a practical (and fun) view.

As Marc Andressen and Bobby Kotick -CEO of Activision Blizzard- recently discussed in the a16z podcast, gaming is becoming mainstream. Since the PUBG launch in 2017, the Battle Royale genre has been growing at a lighting speed. It was followed by the record-breaking Fortnite Battle Royale launch, hitting 10 million downloads in their first two weeks. When no one was expecting it to happen again, Apex Legends launched in 2019 to crush the Fortnite mark with 25 million downloads in its first week. The genre war itself seems to be a Battle Royale game.

We decided to take one week of conversation from Twitter to analyze what people was talking about. What do they like about the new Apex Legends game? What should Fortnite do to go back to the top? Is PUBG still relevant on consoles?. These are some of the questions we had before uploading the dataset to our unsupervised AI platform.

The following insights contain screenshots of the intent clusters discovered, as well as specific examples of documents that were automatically tagged. The top left tag from the documents are manually-created tags that helped organize the findings across the conversation.

© 2019, Electronic Arts Inc.

Apex Legends

Insight 1: The biggest concern is centered around bugs and crashes

“game+crash” cluster from the Apex Legends dataset

The development team behind is already focusing their communication on that topic, as they shared a recent update. In the first weeks they hit 25M players and there is a lot of noise around the issues.

Insight 2: Users are requesting for the Solo & Duo modes

“solo+duo” cluster from the Apex Legends dataset

The game launched only with three-players squad mode. People is actively requesting a Solo and Duo modes, as they are tired of playing “with randoms”. That usually means they play with a teammate, but the third slot is filled up with another player they don’t know. It also happens that solo players are forced to people with others in the same team. It’s worth noting that PUBG and Fortnite have these modes already available.

Document automatically tagged by the Lang.ai intent induction process
© 2019, Epic Games Inc.

Fortnite

Insight 1: Complains that the shop items didn’t change

“same+shop” cluster from the Fortnite dataset

The game changes the items shop in a period manner, but that time it didn’t happen. People was confused about it and some of them got the opportunity to compare it with the recent Apex Legends success.

Insight 2: The community-requested soccer skins are back

“skin+soccer” cluster from the Fortnite dataset

As they recently announced on twitter, they are adding back some skins that people were requesting:

The community is also requesting to bring back the double pump (using two shotguns):

PUBG

© 2019, PUBG Corporation

Insight 1: There is a lot of engagement around PUBG mobile

“mobile+pubg” cluster from the PUBG dataset

The biggest cluster in the PUBG conversation is about PUBG Mobile, probably because they recently launched a Zombie Mode update for their mobile game.

Insight 2: Apex Legends seems to be compared more with PUBG than with Fortnite

“apex+legends” cluster from the PUBG dataset

Fortnite can be classified more like a casual game than PUBG, as we can see with their last-year launch of their creative mode. This may be the reason why there is a stronger pattern of PUBG users moving to Apex Legends, given that Apex shares more similarities with PUBG than with Fortnite.

How we did this

The first step was getting the data we wanted to analyze. We downloaded 20k tweets for each game using the Twitter API and the following query:

fortnitegame OR PlayApex OR pubg

Once we had the dataset ready, we created a project in Lang.ai. Creating the project was as simple as uploading the dataset file and then waiting for the intent induction process to finish. In less than an hour, the results were ready to be reviewed.

If you want to try out Lang.ai, you can request an invite.

Check the other articles in our Building Lang.ai publication. We write about Machine Learning, Software Development, and our Company Culture.

--

--

Technical Product Manager at @_langAI. Open Web Evangelist. Author of “The Success Hacker” book. Get it on http://thesuccesshacker.es.