Playtika AI

Overcoming personalization challenges with AI – Meetup summary

By Michael Kolomenkin

We are happy to share with you the videos of lectures from the DS meetup we organized on June 22nd. The meetup was dedicated to one of the most fascinating questions in the AI realm – how AI can help to make personalized experiences. This is the question that we in Playtika have been working on for a long time.

Today, personalization is more than just a bonus. It is a crucial capability in the customer-centric business landscape. Customers don’t just want personalization. They demand it, especially in the digital world. They got used to it through music and video streaming service and e-commerce recommendation engines. Without the personalized touch, customers are much more likely to churn.

Personalization can be defined as tailoring an experience to an individual based on information a company has learned about the individual. Of course, it is possible to tailor the experience manually when you have tens of customers. But with millions of customers and petabytes of available information, one cannot help using an automatic personalization system.

But building such an automatic, or AI-based personalization system, is very challenging due to two reasons. The first reason is technical. As opposed to the areas where AI has recently obtained great success – image and language processing – the user data is much more difficult to process. It comes from a variety of sources and in a variety of types. Moreover, it is dynamic and “ages” fast. What was true about me half a year ago is not guaranteed to be true today. The second reason is that personalization systems deal with humans. And humans are less predictive than any physical process.

We in Playtika have been working on personalization problems for several years. We applied personalization to churn, in-game recommendations, and other areas. We understand well enough both the challenges and the potential of good solutions.

Therefore, we decided to organize a meetup where we can share our knowledge with the public, learn from other successful companies and ignite the discussion about AI-based personalization.

The videos are:

  • Main challenges in the AI-based personalization of games and entertainment apps by Dr. Michael Kolomenkin. Michael talks about why AI is important for keeping the players engaged, and overviews the main technical difficulties:
  • Online limited memory neural-linear bandits with likelihood matching by Prof. Shie Mannor. She explains his paper, which was published in ICML21:
  • A real-time recommendation system in Playtika games by Dario D’Andrea. Dario describes both the algorithmic and engineering components of the bandits-based system used by Playtika for in-game recommendations:

The craft of matching sea crafts by Dana Averbuch. Dana demonstrates what goes on behind the scenes of Pick a pier – a company that matches seacrafts with ports:

Finally, Alexandra Zhuravlev presents the ideas behind Descriptor. AI. Her talk is about how human emotions can be obtained from sound to improve the user experience:

We hope you enjoy the videos. If you have questions, want to cooperate, or just want to discuss an idea, feel free to write to [email protected]. Stay tuned for the next Playtika meetup on personalization with more new ideas and inspiring presentations.