The Game Recommendation system extensively leverages data from player sessions, considering an array of features including stake amounts, session frequency, preferred games and frequently played titles.
After establishing metrics, our recommendation models typically curate a list of up to 20 game titles per player with designated ranks; games with a higher ranking are more likely to be preferred by players.
If required, the configuration can be adjusted so that a minimum of 20 recommendations is produced per player. In this instance, for players who have less than 20 natural recommendations, we would backfill using recommendations that resonate best with the player base.
Amplifier AI uses the following AI models to deliver real-time or batch recommendations. Notably, the recommendations can be derived independently from each model or combined to form a set of blended recommendations.
- Significant Activity & Favourite Games – The system evaluates each player's gaming session, categorising sessions as 'significant' while filtering out noise from promotional influences and sessions with low engagement. The algorithm determines favourite games by considering factors such as activity, recency, experience, focus and luck.
- Collaborative Filtering – This model considers games a player has engaged with as well as those played by other players who share similar preferences.
- Game Similarity Analysis – Clusters games based on feature similarities such as volatility, RTP, paylines, payout distribution etc. This model identifies similar games to each player’s significant and favourite games.
- Game Sequencing – Predicts player game sequences given the games played so far. This model distinguishes sequences according to player lifecycle and commercial targets, such as conversion for players in the early lifecycle.
The outputs from the models lead to a variety of recommendation types, each catering to different aspects of player preferences. The recommendation system generates the following types of recommendations:
.png?width=688&height=397&name=Rectangle%20158%20(3).png)
If Future Anthem possess data pertaining to specific jurisdictions or sites, this information is factored in when generating recommendations, i.e. only games available in a particular jurisdiction and/or site will be incorporated into the recommendations.
The recommendations are updated on the following basis:
- Players are rescored based upon new activity and/or once their recommendations are 7 days old.
- Recommendations may not necessarily change based on player activity. If there has been new activity but we deem it not to be significant activity, the player will receive an updated recommendation set consistent with the previous one.
- The last updated date (or the date for output file) corresponds to the most recent rescored date.