{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_53314"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_53314","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"LLM-BASED PLAYER-STYLE AGENT GENERATION FOR AUTOMATED GAME PLAYTESTING : MODELING HUMAN PLAY BEHAVIORS TO ENABLE SCALABLE LEVEL EVALUATION","abstract":"Thesis (M.A.)--Michigan State University. Media and Information--Master of Arts, 2026","abstract_html":"Thesis (M.A.)--Michigan State University. Media and Information--Master of Arts, 2026","abstract_has_math":false,"creators":["Zhang, Xunzhuo"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Cote, Amanda","Ratan, Rabindra","Shillair, Ruth"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:47Z","subjects":["Artificial intelligence","Information science","Computer science"],"languages":["eng"],"rights":["In Copyright"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53314","local:Zhang_grad.msu_0128N_22368"],"render_values":[{"text":"etd:53314","href":null,"code":true},{"text":"local:Zhang_grad.msu_0128N_22368","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/2vb8-3v08","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/2vb8-3v08","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_53314","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Cote, Amanda","Ratan, Rabindra","Shillair, Ruth"]},{"key":"dc:creator","label":"Author","values":["Zhang, Xunzhuo"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Artificial intelligence","Information science","Computer science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53314","local:Zhang_grad.msu_0128N_22368","https://doi.org/doi:10.25335/2vb8-3v08"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (M.A.)--Michigan State University. Media and Information--Master of Arts, 2026","As contemporary game design theory places more importance on experience-based design, playtesting serves as an essential stage in digital game development. However, as modern games become larger and more structurally complex, recruiting sufficient real human participants for repeated playtesting is costly, time-consuming, and difficult to scale. Existing automated playtesting systems can reduce part of this cost, but they remain limited in reflecting the behavioral diversity and sub-optimal play patterns of real human players. This thesis examines to what extent large language model (LLM)-based agents can reflect different player types in automated game playtesting.Using ORAK as the experimental platform, this study constructs LLM-based agents conditioned by three player styles based on Bartle’s taxonomy: Achiever, Explorer, and Killer. The experiment is conducted in Super Mario and Slay the Spire using gpt-5.4-mini-2026-03-17. The final dataset includes 150 valid Super Mario traces and 60 valid Slay the Spire traces. Agent-generated play traces are evaluated through behavioral metrics, clustering analysis, and human-anchored comparison with public human gameplay traces. The results show that LLM-based agents can generate distinguishable player-style behaviors to some extent, but this capability is shaped by game environment and player style. In Super Mario, human-anchored comparison reveals weak alignment with real human player types, suggesting a risk of anthropomorphic illusion. In Slay the Spire, alignment is stronger, especially for Killer and Explorer agents, while Achiever remains weaker. Overall, this thesis suggests that LLM-based agents have potential value for automated playtesting, but their behavior should be evaluated through trace-level evidence and human-agent comparison rather than only game outcomes or agent-only clustering.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (114 p.) : ill."]},{"key":"dc:title","label":"Title","values":["LLM-BASED PLAYER-STYLE AGENT GENERATION FOR AUTOMATED GAME PLAYTESTING : MODELING HUMAN PLAY BEHAVIORS TO ENABLE SCALABLE LEVEL EVALUATION"]}]}],"canonical_facts":{"dc:contributor":["Cote, Amanda","Ratan, Rabindra","Shillair, Ruth"],"dc:creator":["Zhang, Xunzhuo"],"dc:date":["2026"],"dc:description":["Thesis (M.A.)--Michigan State University. Media and Information--Master of Arts, 2026","As contemporary game design theory places more importance on experience-based design, playtesting serves as an essential stage in digital game development. However, as modern games become larger and more structurally complex, recruiting sufficient real human participants for repeated playtesting is costly, time-consuming, and difficult to scale. Existing automated playtesting systems can reduce part of this cost, but they remain limited in reflecting the behavioral diversity and sub-optimal play patterns of real human players. This thesis examines to what extent large language model (LLM)-based agents can reflect different player types in automated game playtesting.Using ORAK as the experimental platform, this study constructs LLM-based agents conditioned by three player styles based on Bartle’s taxonomy: Achiever, Explorer, and Killer. The experiment is conducted in Super Mario and Slay the Spire using gpt-5.4-mini-2026-03-17. The final dataset includes 150 valid Super Mario traces and 60 valid Slay the Spire traces. Agent-generated play traces are evaluated through behavioral metrics, clustering analysis, and human-anchored comparison with public human gameplay traces. The results show that LLM-based agents can generate distinguishable player-style behaviors to some extent, but this capability is shaped by game environment and player style. In Super Mario, human-anchored comparison reveals weak alignment with real human player types, suggesting a risk of anthropomorphic illusion. In Slay the Spire, alignment is stronger, especially for Killer and Explorer agents, while Achiever remains weaker. Overall, this thesis suggests that LLM-based agents have potential value for automated playtesting, but their behavior should be evaluated through trace-level evidence and human-agent comparison rather than only game outcomes or agent-only clustering.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (114 p.) : ill."],"dc:identifier":["etd:53314","local:Zhang_grad.msu_0128N_22368","https://doi.org/doi:10.25335/2vb8-3v08"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["In Copyright"],"dc:subject":["Artificial intelligence","Information science","Computer science"],"dc:title":["LLM-BASED PLAYER-STYLE AGENT GENERATION FOR AUTOMATED GAME PLAYTESTING : MODELING HUMAN PLAY BEHAVIORS TO ENABLE SCALABLE LEVEL EVALUATION"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:47Z"}