Data Science @ Activision

Two data scientists from Activision gave a talk at @crosscampusLA last night for the Machine Learning Meetup.

Download the audio here (make take several minutes due to my hosting site).

Summary:

The first speaker was Josh Hemann on the Game Analytics team. The team’s goal was to make the game experience better.

He called one of the improvements: “Vinette 1: Algorithm Detection of Assholes.” Some players found a shortcut to raising their rank in Call of Duty known as “boosting.” Two players enter a game on opposite teams and take turns killing each other. This quickly improves their rank. At first, most of the “boosting” detection had to be done manually with Activision players surveying random games. However, the team used algorithms and data patterns to determine how quickly the player jumped in rank, the number of times they killed the same player, the number of games with the same player, the player positions on the map, etc. The number of cheaters detected jumped to 200 per day.

The team’s second improvement came from customer feedback and was called “Vignette 2: Natural Language Processing.” Activision doesn’t have data immediately after a game launch. Instead, they run algorithms against social sites such as Reddit to learn about their audience’s feedback. Unfortunately, the beta and pheta algorithm explanation lost my attention until the next speaker took the stage.

Mae Coughlin, a data scientist on the Marketing and Advanced Analytics team, started the presentation with a song and some visuals. Coughlin showed how their marketing data affected sales decisions. For example, graphs included time played vs. retention or churn vs. amount of DLC purchased. The data was so specific that Activision could estimate the monetary value for one hour of play for one person.

The team is also working on other projects such as a personalized 1:1 targeting email strategy to reduce churn. Novice players receive encouragements, while expert players learn about new DLC or “Double XP” weekends. Then, the team can see if those who received emails decided to continue or increase playing COD. Currently, the number of unique email templates ranges in the 6-figure mark.

Coughlin closed the talk by encouraging data scientists to consider Activision for employment.


Originally published at veerkampvisuals.com.

Show your support

Clapping shows how much you appreciated Shannon Veerkamp’s story.