Uz emocionālu stāvokli balstīta adaptīva stāstu veidošana ar AI

Abstract

Adaptīvā stāstu veidošana pēdējās desmitgadēs ir ievērojami attīstījusies. Šī pieeja ir piesaistījusi pētnieku uzmanību arī videospēļu jomā. Ir pieaudzis spēļu spēlētāju skaits un daudzveidība (gan pēc demogrāfijas un izcelsmes, gan personības aspektiem). Parasti spēļu izstrādātāji balsta savus stāstus uz savas pašu izpratnes par auditorijas vēlmēm. Tomēr tas negarantē, ka ikkatram spēlētājam spēles gaita šķitīs apmierinoša. Savukārt auditorijas pašreizējā noskaņojuma atpazīšana ļauj pasniegt stāstu no unikāli personiskas perspektīvas. Šī darba autors piedāvā jaunu pieeju personalizētu interaktīvu sižetu veidošanai, pamatojoties uz spēlētāju sejas emociju izpausmēm videospēles laikā. Autors pieļauj, ka spēlētāji vairāk iesaistās un izbauda videospēli, ja spēles stāstījums tiek vadīts un dinamiski adaptēts atkārībā no spēlētāju emocijām. Autors izveidoja videospēles prototipu, kas atpazīst, vai spēlētājs ir skumjš, neitrāls vai pārsteigts, un attiecīgi pielāgo spēles stāstījumu. Lai izmērītu spēlētāju apmierinātības līmeni, autors izveidoja arī prototipa otru versiju, kurā netiek pielietota spēles notikumu adaptācija. Autors, veicot aptauju, salīdzināja spēlētāju apmierinātības un iesaistīšanās līmeni pēc abu spēļu spēlēšanas. Aptaujas rezultāti pierādīja, ka spēlētāji vairāk izbauda tādu spēli, kas pielāgojas viņu emocionālajam stāvoklim. Atslēgvārdi: adaptīvā stāstīšana, sejas izteiksmes, emociju atpazīšana, konvolucionālais neironu tīkls. Diplomdarbs

In recent decades, adaptive storytelling has been developing considerably. Recognizing the audience’s current state of mind allows to give them a story from a uniquely personal perspective. This approach especially attracted the attention of researchers in the video game field. The number of game players and their diversification have increased. Usually, game developers base their stories only on their own understanding of the average preferences of their audience. However, this does not guarantee satisfying storytelling for every player. This is why the author of this thesis proposes a new approach for creating personalized interactive storylines based on the facial emotion expressions of users during the video game. The author created a prototype of the video game that recognizes whether the player is sad, neutral, or surprised and adjusts the game's narrative accordingly. The author used two Convolutional Neural Networks models to detect players’ emotions. To measure the level of enjoyment of the players, the author created the same game but without adaptation element. The author conducted a survey to compare the level of enjoyment and engagement of players after playing these two games. The results of the survey proved that players feel more enjoyment when playing a game that adapts to their emotional state. The created game model and gained results have considerable benefits. They open a new direction of video game industry development. The created prototype demonstrates the method of creating interactive and more enjoyable game experience that can attract audience’s attention and interest. However, the author faced several limitations that could be considered in further research to get more precise results. The main limitation is related to the CNN models. The accuracy of the models that were used in this work depended significantly on several factors, such as the level of lighting in the player’s room. The models could be trained better to detect emotions more accurately. Furthermore, it is important to mention that the accuracy of game adaptation depended on how vividly players showed their emotions through their facial expressions. This is why it was harder for a model to recognize emotions correctly of a player who was calmer and not expressive. The author of the paper states that there is a space for future research by using this work’s method. It could be useful to implement a bigger spectrum of emotions that get recognized, for instance, happiness, fear, or disgust. This will also allow to create more narrative branching and create more engaging storylines. The authors of future works could create various game environments and include different types of game activities. For example, if a player does not like fighting with enemies, then he/she can play, for instance, more intellectual games, such as puzzles, that are parts of the game’s storyline. The thesis consists of 46 pages and includes 27 sources, 5 tables, 7 figures, as well as 2 Appendices with 1 table and 1 figure. The keywords of this work are adaptive storytelling, facial expression, emotion recognition, convolutional neural network.