Targeted Upskilling Framework based on Player Mistake Context in Online Skill Gaming Platforms

Abstract

Recent proliferation of online platforms for Games of skill have made—an excellent source of recreation and relaxation, and the safest avenues for realising personal worth, respect, and recognition— readily accessible on the tip of a finger. A primary driver towards this enjoyment is the skill of a person in responding to complex game dynamics and states, which eventually impacts the game outcomes. Hence, from both player and platform perspectives, improving the player skills—or conversely, reducing player mistakes—is paramount to improve player experience and engagement. In this talk, we focus on unique personalized and near-real-time up-skilling framework depending on specific mistake contexts identified for the players. This framework leverages a suite of models to determine the correct action in a given game state. For a specific case-study of Rummy—a popular skill game in India—these are CNN-driven deep learning models. However, our framework can be plugged with any other suite, action set, and game state representations (depending on the game) within a broad construct of getting the best reference action set in a game state. Players’ actions are then benchmarked w.r.t. the reference actions as adherence measures. A lower adherence essentially indicates deviations from correct actions, i.e., the mistakes. An explainable set of rules are accordingly derived from game features, which sets the contexts for these mistakes, leading to opportunities for targeted up-skilling.