Game recommendation Engine - In app purchase

recommendation_engin
machine_learning
logistic_regression

#1

Hello,

I am trying to build a recommendation engine for a popular mobile game(ios and android). Currently the app doesn’t have a recommendation engine that is data driven. All the current recommendation engine are based on clicks and are generic.

I have around 5 million user records where only 5% of them have ever made a purchase. How do i go about building a recommendation engine for such case.

Data:
I have click events of a user session like game starting, ending, click on store item etc etc. I have user country and device information and that’s all

Please advice

Thanks


#2

Could you please clarify whether you are working for a personalized recommender or generic?

PS : Good resource for choosing a recommender system


#3

@Amit_Sood

Interesting question and application. While I am not closely involved in gaming industry, here are a few case studies I have heard in past. I also don’t know whether 5% is good or bad from industry benchmark - it might make sense to check that as well.

Some gaming companies are using more granular game experience data to pitch in app purchases. For example, if you are having a bad day at Poker, you will get a recommendation to watch an advertisement to get a few chips. On the other hand, if your day day is going well - you would probably not watch an advertisement. Once the user has seen advertisement / shown intent to make in app purchase, you can then provide them an offer, which should result in better conversion.

So, in essence, if you use the entire gaming experience in a session and compare it to the user history and other user history, it can start throwing you more ideas about right ways to pitch in app purchase.

Hope that helps.

Regards,
Kunal


#4

This is going to be a generic recommendation engine.


#5

@kunal, Appreciate the reply. Thank you

Industry benchmark is around 8%. This 5% is not necessarily from a recommendation. The game in its current state do have certain triggers on when to show an in app purchase, like(its a fighting game) so if a player is running out of power then it triggers a recommendation to either buy or watch a video to get more power.

But these triggers are hard coded and are not driven from data intelligence.

Being a gamer myself some of the basic triggers that come to my mind are:

  1. Being stuck at a certain level for some time(you mentioned this as well).
  2. How often do i like to collect the coins and how(either through watching adverts. or spending hard cash).
  3. Do i often visit in app store(may or may not make a purchase). May be i am looking for a good offer daily.

My trouble is how do i model this. My data has only click events. What techniques should i use.