Recommendation systems for influencing player competence and engagement in match-based games

Abstract

For years, video games have garnered the attention of people worldwide, providing a unique environment for entertainment, education, scientific problem solving, promotion of health awareness, and socialization. With the increase of video games' popularity, a critical question arises: how do we keep a player engaged? In this dissertation, I focused on increasing player engagement, where I define engagement as the continued desire to play a game repeatedly during one session or over a longer period of time. I proposed three recommendation systems targeting different types of in-game elements. I focused on online match-based games, where players competitively play against others. Further, I focused on recommendations in the pre-match stage. My hypothesis is that the recommendation systems I developed will increase player engagement by giving players the ability to make informed choices concerning in-game elements, such as which characters to play as, which items to equip, or which opponents to play against. My theoretical foundation for this work is based on two psychological theories that have shaped understanding of player sustained engagement in games: Self Determination Theory and Flow Theory. In such theories, scholars have shown that competence, or the ability of a player to feel competent given the tasks they are doing within the game, has a direct relation to sustained engagement, or the desire to play the game repeatedly. Additionally, Flow theory describes a state of 'flow' where players are sufficiently challenged, where the task they are doing is neither too easy or too hard. Based on these theories, the recommendation systems I proposed recommend in-game elements to players that influence their in-game choices, and subsequently will have an impact on the way they play, competence they perceive, and eventually their engagement. To develop these recommendation systems, I explored two related research questions. The first is concerned with how to develop effective and efficient recommendation systems that can recommend in-game elements that will have a winning potential for players, in both one-vs-one and team-vs-team settings. Targeting this question allows us to focus on developing experiences that are not frustratingly hard for players, and thus help increase their winnings. It should be noted that these systems can also work in reverse, meaning they can also suggest items that make players lose. The second research question is concerned with studying how to balance win/loss ratio of matches through designing in-game element recommendation systems which directly improve player engagement in relation to win/loss ratio and its impact on the churn rate. For the first research question I proposed two systems: (a) Q-DeckRec, a recommendation system for one-vs-one Collectible Card Games (CCGs). In these types of games, the player often starts with a deck of cards of their choice. Q-DeckRec recommends winning-effective decks. This system presents a novel approach to search a large space of possible card decks to recommend, using minimal computational resources after a training phase. In addition to cards as starting items to recommend, many games, such as MOBA (Multi-player Online Battle Arena games, require players to select a set of characters to play with against another team. For these types of games, I proposed (b) DraftArtist, a recommendation system that recommends winning-effective characters. This proposed system presents a novel contribution that efficiently searches for the best possible characters to recommend given a large number of possibilities, uncertainty caused by not knowing what the other team will select, and the desirability to select characters that synergize with teammates and counter the opponents. Using a match outcome prediction model trained on real data, we find that teams following our recommendation algorithm have higher predicted win rates against teams constructed by other character selection strategies. Our algorithm maintains sufficient efficiency to be deployed in real-world scenarios. For the second research question, I proposed an opponent recommendation system, called Engagement-Optimized Matchmaking (EOMM). EOMM is the first system that formally treats matchmaking as an optimization problem to maximize player engagement quantitatively. Our system shows significant improvement in enhancing player engagement across all players as compared to other matchmaking methods. The contribution of my dissertation is in developing three novel recommendation systems that target improving player engagement through increasing competence by giving players in-game element recommendations that increase possibility of a players feeling competent or entering a state of Flow by either increasing the probability of them winning or balancing the win/loss ratio to decrease the probability that they leave the game. There are much work left for future research, such as studying systems tailored for more specific types of in-game elements, as well as those which can make central decisions on different types of recommendations for any particular player.