20.500.12556/DKUM-97442 Vpliv lokalnih in oblačnih inteligentnih neigralnih likov na uporabniško izkušnjo v vr igrah magistrsko delo The impact of local and cloud based intelligent non player characters on the user experience in virtual reality Interakcije z neigralnimi liki v sodobnih videoigrah in okoljih virtualne resničnosti kljub tehnološkemu napredku pogosto ne dosegajo pričakovanj uporabnikov. Vedenje likov je nerealistično in predvidljivo. Magistrska naloga to naslavlja s primerjavo dveh pristopov k implementaciji inteligentnega neigralnega lika z govorno interakcijo. Lokalni pristop za razumevanje igralčevih ukazov uporablja model strojnega učenja za zaznavanje semantične podobnosti stavkov, oblačni pristop pa prek storitve OpenAI API uporablja velike jezikovne modele. Obe različici sta bili implementirani v obliki inteligentnega asistenta v prototipni igri VR, postavljeni v okolje restavracije, ter ovrednoteni z znotraj-osebnim eksperimentom s 19 udeleženci. Merjeni so bili zaznana prepričljivost vedenja, socialna prisotnost, odzivnost lika ter prisotnost. Analiza rezultatov je pokazala, da oblačni pristop statistično značilno prispeva k višji zaznani prepričljivosti, pri preostalih pa med pristopoma niso bile zaznane statistično značilne razlike. Despite technological advances, interactions with non-player characters in contemporary video games and virtual reality environments frequently fall short of user expectations. Character behaviour remains predictable, and responses are often limited and unnatural. This master's thesis addresses this shortcoming by comparing two approaches to implementing an intelligent non-player character capable of voice interaction. The local approach employs a machine learning model for detecting semantic sentence similarity to interpret player utterances, while the cloud-based approach leverages large language models via the OpenAI API. Both versions were implemented as an intelligent assistant in a prototype VR game set in a restaurant environment and evaluated through a within-subjects experiment involving 19 participants. The measured constructs included perceived behavioural believability, social presence, and responsiveness of the character. Analysis of the results revealed that the cloud-based approach led to significantly higher perceived believability, while no statistically significant differences between the two approaches were found for the remaining constructs. Virtualna resničnost inteligentni neigralni lik govorna interakcija uporabniška izkušnja prepričljivost virtualnega agenta veliki jezikovni modeli semantična podobnost stavkov lokalni model strojnega učenja Virtual reality intelligent non-player character speech interaction user experience virtual agent believability speech recognition large language models semantic sentence similarity local machine learning model true true false [B. Čerpnjak] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2026-03-07 16:41:32 2026-05-08 13:34:30 2026-05-16 03:14:20 0000-00-00 00:00:00 2026 0 Maribor 0 Maribor 1 spletni vir (1 datoteka PDF (XII, 122 str.)) 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.946(043.2) 278317827 MAG_Cerpnjak_Blaz_2026.pdf MAG_Cerpnjak_Blaz_2026.pdf 1 CC84E2F81E46E9363F5F067A234793F1 fef026b9629352ba1885963097c99d3c3ffc3cc7e4c0715243b1cd60ffbf0d35 a249dd5b-1ad4-11f1-b1f6-00155d580e1a https://dk.um.si/Dokument.php?lang=slv&id=204359 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0