Twelfth International Workshop Modelling and Reasoning in Context (MRC) @IJCAI 2021 51 Bartle Taxonomy-based Game for Affective and Personality Computing Research Laura Żuchowska1 , Krzysztof Kutt2∗ and Grzegorz J. Nalepa1,2 1 Department of Applied Computer Science, AGH University of Science and Technology, Kraków, Poland 2 Jagiellonian Human-Centered Artificial Intelligence Laboratory (JAHCAI) and Institute of Applied Computer Science, Jagiellonian University, Kraków, Poland krzysztof.kutt@uj.edu.pl, gjn@gjn.re Abstract 2021a]. In the latter case, games are used as a fully control- lable experimental environment that allows for accurate mon- The paper presents the design of a game that will itoring of the user’s interaction with the system [Żuchowska serve as a research environment in the BIRAFFE et al., 2020]. This is possible due to the similarities in human- series experiment planned for autumn 2021, which in-the-loop [Nunes et al., 2015] and affective loop interaction uses affective and personality computing methods schemes. However, in order to extend the results of AfG re- to develop methods for interacting with intelligent search to interaction models of intelligent assistants in the fu- assistants. A key aspect is grounding the game de- ture, careful game design and a system for logging the whole sign on the taxonomy of player types designed by game context are required [Kutt et al., 2021b]. Bartle. This will allow for an investigation of hy- potheses concerning the characteristics of particu- The notion of context is understood as a component of lar types of players or their stability in response to emotion, according to the theory proposed by Prinz [2006]. emotionally-charged stimuli occurring during the In this view, context is anything that allows one to interpret game. a particular physiological activation and give it an appropri- ate interpretation. In the BIRAFFE series of experiments, the primary contextual information is behavioral data describing 1 Introduction and Motivation the interaction with the system/game – both the user’s ac- Affective Gaming (AfG) [Lara-Cabrera and Camacho, 2019] tions and the stimuli appearing in the system/game. In ad- is an area of research concerned with how games can measure dition, demographic information (gender, age) and personal- and detect player emotions, and then use this information to ity profiles are collected. Ultimately—when we move from adapt the game environment accordingly. If these modifica- a game-based experimental environment to real-world intel- tions are also aimed at directing the player’s affective state, ligent assistants—external sources of context, e.g., calendar e.g., towards specific emotions desired at a given stage of the data, current weather, will also be used. Importantly, once we game, then one can call this an affective feedback loop in have refined the low-level context storage mechanisms de- which the game and the player interact (see Fig. 1). scribed in this paper, we also plan to attempt to derive higher- level context from them, e.g., instead of relying on changes in the position of individual characters in the game, we will operate on the information “the player is attacking an enemy” or “the player is running away from an enemy” instead. This paper summarises the work carried out to prepare the game for the third experiment in the BIRAFFE (Bio- Reactions and Faces for Emotion-based Personalization) se- ries. The motivation for developing the game in question was twofold. The first intention was to improve the experimen- tal environment based on lessons learned from previous stud- ies [Kutt et al., 2021a; Kutt et al., 2021b], in particular to pro- vide a more accurate game context logging system. The sec- Figure 1: Affective game feedback loop [Lara-Cabrera and Cama- cho, 2019]. ond motivation was to extend the game design to include dif- ferent types of interaction for different types of players. Com- bining information on the said types with personality profiles Studies in the AfG area do not just focus on entertainment. and physiological characteristics—obtained in all BIRAFFE It can also be part of research projects concerning educa- experiments—will enable broader analyses that could lead to tion [Dormann et al., 2013] or the design of intelligent assis- the identification of a set of characteristics for each type of tants, as in the BIRAFFE series of experiments [Kutt et al., player. It will also allow to investigate if and how the types ∗ Corresponding Author and characteristics of users change during the course of the Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). Twelfth International Workshop Modelling and Reasoning in Context (MRC) @IJCAI 2021 52 game. some minor exploration can be included, in order to under- The rest of the paper is organized as follows. In Sect. 2, stand what other person is referring to. The act of killing is Bartle taxonomy of player types is introduced. The design of not required nor wanted for such a person to have fun. For the game with multiple paths for all player types is discussed single-player games, the socializer can be more entertained in Sect. 3. Then, in Sect. 4 the set of logged contextual infor- by making interesting NPC’s with interesting backstory, mul- mation is described. The paper is concluded in Sect. 5. tiple dialogue options and arcs, thought-provoking and en- gaging plot can be enough for socializer too. 2 Bartle Taxonomy Killers are a very specific, narrow group of people. Inten- tions behind a mind of a killer are clear to some extent. They The Bartle taxonomy [Bartle, 1997] is created on a 2D space, enjoy being superior and high in hierarchy, however this is where the X axis is described as “Player – World”, meaning not in the same nature as achievers. Killers tend to do things the involvement of real people instead of non-playable char- they wouldn’t normally do in real life, varying from punch- acters (NPC) or world exploring in any way possible. The Y ing a person to brutal murder. They also cherish the fact that axis is set as “Acting – Interacting”, which directly implicates they can do something to real human, who feels emotions the preference for acting or interacting. Each quarter of the and reacts, instead of NPC. Enjoyment comes from acting on space defines a different type of player as presented on Fig. 2. people. Killers see other people, especially achievers, who can face the challenge, as their prey. Acting 3 Game Design with Multiple Paths for Bartle’s Player Types Killers Achievers The main goal of the game is to use the knowledge of player types in order to get closer to creation of a truly affective ex- perience. As the BIRAFFE3 experiment is aimed to check Players World the associations between the gamer’s personality traits, phys- iological characteristics and in-game decisions (as introduced in Sect. 1), the proposed game provides an open world with as much non-linearity as possible [Gary, 2018]. The affectiv- Socializers Explorers ity of the game has also been taken into account in the de- sign – important choices will be accompanied by emotionally evocative stimuli, both sounds and images. Interacting This game is fairly different than previous ones [Kutt et al., 2021a; Żuchowska et al., 2020], as it provides a pleas- Figure 2: Bartle’s taxonomy of player types [source: https://en. ant gaming experience – something for everyone, no matter wikipedia.org/wiki/Bartle_taxonomy_of_player_types]. if a skilled player or casual person with no gaming back- ground. Multiple point-increasing interaction systems have The achievers’ goal is to act within the world. They wish been introduced, such as dialogues with in-game characters to master the game, find the best possible weapon, get all the (see Fig. 3). The story of each character is very simple, but points (or achievements). People who take care about ranking rewarding enough to keep it entertaining for a subject [Torta and hierarchy can be considered achievers, therefore every and Minuty, 2017]. Some tasks and quests can be done for competitive player is most likely an achiever. Another famous NPCs, mostly in a cute-bubbly way. In order to achieve that, concept for achievers playing type is grinding – playing a all interactable, pickable objects have a type – consumable, game as long as it requires to get a desired outcome [Hilgard plot, weapon or non-consumable. The next important inter- et al., 2013]. action type is attack, which allows to kill an NPC or an animal Explorers start at simple exploring a topology of a game in game with a previously found and equipped weapon item. (breadth) and end at breaking the laws of in-game physics It is important to notice that there is no difference in points (depth), searching and using bugs. They are interested in in- added, whether the action is peaceful or not, the outcome in teracting with the world. This player type is searching for terms of points is always the same. knowledge and likes to be praised by others for having it. The sole purpose of the aforementioned affective pictures While game glitches are fun, players who like to find a spe- and sounds is to induce certain emotions in players, and see cific, unique places and interesting features are also consid- their reactions – the images and sounds will be displayed af- ered explorers. Additionally, speed-runners can also be la- ter some actions have been made. One of the most important beled as a mix between achievers (if ranking is involved) and activities, resulting in revealing a questionable image and/or explorers. sound, is chest opening. Opening such a special chest is one For socializers, the most important part of the game is of many ways to gather points, however there is a trick to community and people, relations with them and interactions. it. There are three types of chests: one with pleasant sounds They love to talk, sympathize and joke with others and appre- and images, second one with 50:50 ratio to get a pleasant or ciate the significance of interacting with players. For some disgusting image, and the third one which always displays socializers, observing the gameplay is enough. For others, an unpleasant, gore image. Every chest varies in terms of Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). Twelfth International Workshop Modelling and Reasoning in Context (MRC) @IJCAI 2021 53 Figure 5: Intentionally placed “bug” in collider, allowing to get out of the map. Figure 3: Dialogue with the NPC. Figure 6: Hidden board placed out of the map as an “easter egg”. sibility to get out of the map (see Fig. 5) and find a hidden Figure 4: Information about completed achievement. board with a nice message written on it (see Fig. 6). As for socializers, NPC are introduced, with their own backstories and problems to solve. Action with an NPC triggers an UI amount of points it gives, which may result in an interesting with dialogue options (see Fig. 3), allowing to know the char- insight on the subject’s importance of points and horrible im- acter better and have a conversation. Killers can find pleasure age watching. Of course, some people might not be interested in killing everybody around and committing acts that would in gathering points in the first place, which creates a challenge be considered illegal or immoral in real life. to overcome, as the images are a crucial part of affective ex- Technically, according to the assumptions made, the game- perience. As far as Bartle taxonomy is considered, all types play time should last 15 minutes. After that time, the game of gamers will find a way to see an affective image and hear a will end and proceed with the experimental procedure (as in sound on a regular basis during the gameplay. Another ways other BIRAFFE experiments, see, e.g., [Kutt et al., 2021a]). to get the subject to look at such a picture include display- There is no possibility to finish game earlier, however there is ing an UI interface by talking with non-playable characters nothing that keeps the subject from just standing in place for or reading boards and interacting with objects. After some 15 minutes and stare blankly at the screen. The whole game random number of lines of text has been displayed, an image was developed with the Unity Engine (https://unity.com/). will be displayed in the background, however there will be no points for that, and the image will be random. Additionally, 4 Logging System when achievement is unlocked by the player, depending on To conduct a study based on such game, a suitable log its type, a pleasant or undesirable sound will be played. handling had to be added. Similarly to the previous re- The whole game design was made specifically with a view search [Kutt et al., 2020], logs are created for each subject, to pursue the characteristics of each player type from Bartle based on their ID defined at the beginning of the experiment. taxonomy. Achievers can find multiple weapons and gather A proper directory is created, along with all files about the points, look up into current statistics and collect achievements game. During the gameplay, data containing current state of for certain actions. The amount of points gathered thorough the player an the progress is being gathered with 10 Hz fre- the game is being shown all the time in top left corner of the quency. A log with an ID of subject as the name is written screen. Achievements on the other hand, are only displayed into JSON file and is being saved in application persistent with the moment of completion (see Fig. 4). The first achieve- data path. Such a log consists of various information about ment will be very simple, in order to show that achievement current state of the game: gathering is possible, triggering some emotions in subjects with particular tendencies. Explorers will be interested by 1. Timestamp, searching for hidden objects on the map and exploiting the 2. Location – both X and Y coordinates and area, mechanics, as some places have intentionally placed “bugs” as easter eggs. One of those bugs is an askew collider for 3. List of unlocked achievements, map. In the bottom left corner of the game, there is a pos- 4. Amount of interaction button clicks, Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). Twelfth International Workshop Modelling and Reasoning in Context (MRC) @IJCAI 2021 54 for the killers, the data on the amount of NPC killed and type of equipped weapon is logged. 5 Summary and Future Work The BIRAFFE series of experiments, which has been run- ning for several years, focuses on the development of interac- tion models for personalised intelligent assistants based on a range of contextual information about the user: physiological Figure 7: Trigger colliders for area logging information. signals collected with low-cost wearable devices, personality assessment, behavioural data describing the interaction with 5. Number of interactions with unique objects, the system, and external sources of context (such as current weather conditions). A means to the goal is to use games as 6. List of particular milestones for NPCs tasks and dia- a stimulus-rich yet fully controllable experimental environ- logues, ment. 7. If talking – name of the NPC, else an empty string, This paper presents the design of a new affective game to be used in the BIRAFFE3 experiment, scheduled for au- 8. Amount of killed NPCs, tumn 2021. In addition to addressing the weaknesses found 9. Current equipped weapon, in previous games, a new contribution of using Bartle’s taxon- omy during interaction design is introduced. This will enable 10. Points and health, post-experimental analyses focusing on determining the char- 11. List of items gathered, acteristics of specific user types or investigating the stabil- ity/variability of player type in response to positive/negative 12. List of opened chests, stimuli associated with their in-game interactions. We be- 13. ID of played sound and image. lieve that inclusion of Bartle player types into both the design The log file can be separated into groups. The first two of the affective game, as well as data analysis about player items (items 1-2 on the list) are purely about the position over interaction with it, provides a new and important source of time of the protagonist, which may help with visualization context. or classification of commonly walked places in game. The Finally, the post-experimental analyses will also focus on “area” is a term describing important places in the world iden- creating a catalogue of interaction patterns, which will be the tified by arbitrarily prepared colliders (see Fig. 7). Second basis for creating an improved version of the game, allowing group (items 3-12) contains the characteristics of players be- the gameplay to adapt to the player’s emotions, i.e., imple- havior – did the protagonist gather achievements? Was s/he menting a full affective game feedback loop. This will thus talking with NPCs? Maybe the subject was killing them? If allow a transition from a “Detection and measure” approach so, with which weapon? How many points were gathered, to an “Integral approach” according to the Lara-Cabrera and etc. This section of logging system is supposed to help in Camacho’s taxonomy [2019]. analysis the most, as the heart of information about a pattern of playing. The last item (13) is for affect-related analyses – Acknowledgements the ID of sound and image displayed after event. The research has been supported by a grant from the Prior- Another log file contains the data about current state of the ity Research Area Digiworld under the Strategic Programme world. The characters are moving all the time, therefore their Excellence Initiative at the Jagiellonian University. location needs to be written down as well – the position of each character can have an impact on each gameplay. References Finally, the last file, which is the same for all players, is the static map of the game world. It consists of information [Bartle, 1997] Richard Bartle. Hearts, clubs, diamonds, about the starting position of items, colliders, houses, etc. It’s spades: Players who suit muds. The Journal of Virtual purpose is to allow for possible future visualization of events Environments, 1(1), 1997. and analysis of collider interactions between the player and [Dormann et al., 2013] Claire Dormann, Jennifer R Whit- the world. son, and Max Neuvians. Once more with feeling: Keeping the Bartle taxonomy in mind, the log can be Game design patterns for learning in the affective domain. also separated into items related to specific gamer types. In Games and Culture, 8(4):215–237, 2013. terms of achievers, the information about points gathered and [Gary, 2018] Justin Gary. Think Like a Game Designer. achievements unlocked is written, along with particular mile- stones for NPC’s quests. The latter can also be used as a Aviva Publishing, Lake Placid, NY, 2018. socializer trait, which is why the data on dialogue options [Hilgard et al., 2013] Joseph Hilgard, Christopher Engel- clicked is also being saved – who was the player talking to. hardt, and Bruce Bartholow. Individual differences in mo- As for the explorers, the amount of unique objects interacted tives, preferences, and pathology in video games: the gam- with together with the amount of interaction button clicks, ing attitudes, motives, and experiences scales (GAMES). chests opened and list of items is written into the file. Finally, Frontiers in Psychology, 4:608, 2013. Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). Twelfth International Workshop Modelling and Reasoning in Context (MRC) @IJCAI 2021 55 [Kutt et al., 2020] Krzysztof Kutt, Dominika Dra˛żyk, Maciej Szela˛żek, Szymon Bobek, and Grzegorz J. Nalepa. The BIRAFFE2 experiment. study in bio-reactions and faces for emotion-based personalization for AI systems. CoRR, abs/2007.15048, 2020. [Kutt et al., 2021a] Krzysztof Kutt, Dominika Dra˛żyk, Szy- mon Bobek, and Grzegorz J. Nalepa. Personality-based affective adaptation methods for intelligent systems. Sen- sors, 21(1):163, 2021. [Kutt et al., 2021b] Krzysztof Kutt, Laura Żuchowska, Szy- mon Bobek, and Grzegorz J. Nalepa. People in the con- text – an analysis of game-based experimental protocol. In MRC@IJCAI 2021, 2021. in press. [Lara-Cabrera and Camacho, 2019] Raúl Lara-Cabrera and David Camacho. A taxonomy and state of the art revision on affective games. Future Generation Computer Systems, 92:516–525, 2019. [Nunes et al., 2015] David Sousa Sousa Nunes, Pei Zhang, and Jorge Sá Silva. A survey on human-in-the-loop appli- cations towards an internet of all. IEEE Commun. Surv. Tutorials, 17(2):944–965, 2015. [Prinz, 2006] Jesse J. Prinz. Gut Reactions. A Perceptual Theory of Emotion. Oxford University Press, Oxford, 2006. [Torta and Minuty, 2017] Stephanie Torta and Vladimir Minuty. Storyboarding: Turning Script into Motion. Mercury Learning and Information, Dulles, VA, 2017. [Żuchowska et al., 2020] Laura Żuchowska, Krzysztof Kutt, Krzysztof Geleta, Szymon Bobek, and Grzegorz J. Nalepa. Affective games provide controlable context. proposal of an experimental framework. In Jörg Cassens, Rebekah Wegener, and Anders Kofod-Petersen, editors, Proceed- ings of the Eleventh International Workshop Modelling and Reasoning in Context co-located with the 24th Euro- pean Conference on Artificial Intelligence, MRC@ECAI 2020, Santiago de Compostela, Galicia, Spain, August 29, 2020, volume 2787 of CEUR Workshop Proceedings, pages 45–50. CEUR-WS.org, 2020. Copyright c 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).