Personalized Non-Player Characters: A Framework for Character-Consistent Dialogue Generation
Open accessThe DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index.
Abstract
Generating character-consistent and personalized dialogue for Non-Player Characters (NPCs) in Role-Playing Games (RPGs) poses significant challenges, especially due to limited memory retention and inconsistent character representation. This paper proposes a framework for generating personalized dialogues based on character-specific knowledge. By combining static knowledge fine-tuning and dynamic knowledge graph technology, the framework generates dialogue content that is more aligned with character settings and is highly personalized. Specifically, the paper introduces a protective static knowledge fine-tuning approach to ensure that the language model does not generate content beyond the character’s cognitive scope during conversations. Additionally, dynamic knowledge graphs are employed to store and update the interaction history between NPCs and players, forming unique “experience-response” patterns. During dialogue generation, the paper first parses player input into an Abstract Meaning Representation (AMR) graph, retrieves relevant memory nodes from the knowledge graph, and constructs a fused graph structure. This integrated graph is encoded via a graph neural network to generate high-dimensional semantic vectors, which are then used to retrieve and supplement knowledge from the vector database. Ultimately, the model generates personalized responses consistent with the NPC’s identity. Experimental results demonstrate that the framework significantly enhances the authenticity of NPC dialogues and player immersion and performs well on multiple large-scale language models.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Prior work demonstrates that LLMs can synthesize characters, quests, narratives, and dialogue with high linguistic fluency and diversity, supporting applications ranging from conceptual game design to LLM-driven NPCs and open-ended simulations [1-4].”
From Game Knowledge Management System: Schema-Governed LLM Pipeline for Executable Narrative Generation in RPGs (2026) · cites this paper as [3]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Glove: Global Vectors for Word Representation
Pennington, J., Socher, R., & Manning, C. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1162 · Cited by 34,071 · cited 1×
“It primarily initializes the concept node vectors in the AMR–KG graph using GloVe [35] word embeddings and absolute position embeddings.”
Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A., Levy, O., Stoyanov, V., & Zettlemoyer, L. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.703 · Cited by 8,821
Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A., Levy, O., Stoyanov, V., & Zettlemoyer, L. (2019) · arXiv · doi:10.48550/arxiv.1910.13461 · Cited by 8,816
Human Memory: A Proposed System and its Control Processes
Atkinson, R. C., & Shiffrin, R. M. (1968) · Psychology of Learning and Motivation · doi:10.1016/s0079-7421(08)60422-3 · Cited by 7,216 · cited 1×
“To address this, this study draws on the Atkinson–Shiffrin [7] model of human memory to construct an NPC memory management system composed of short-term working memory and long-term semantic memory.”
Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., et al. (2025) · ACM Transactions on Information Systems · doi:10.1145/3703155 · Cited by 4,688 · cited 1×
“…research mainly relies on the problem recognition capability of Large Language Models (LLMs), but traditional sequence-to-sequence architectures are prone to introducing noise and, influenced by the “hallucination” problem [10], may lead to relationship recognition errors or entity omissions.”
LLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602 · cited 1×
“During the training phase, we selected the LLaMA 7B model [41] as the base model and trained it using a supervised fine-tuning strategy.”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3586183.3606763 · Cited by 2,187 · cited 1×
“Teams from Stanford University and DeepMind [18] demonstrated the ability of LLMs to simulate human memory, reflection, and dialogue through prompts.”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · arXiv · doi:10.48550/arxiv.2304.03442 · Cited by 2,173 · cited 1×
“Teams from Stanford University and DeepMind [18] demonstrated the ability of LLMs to simulate human memory, reflection, and dialogue through prompts.”
Malkov, Y. A., & Yashunin, D. A. (2020) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2018.2889473 · Cited by 1,904 · cited 1×
“Vector databases store data [25] in the form of high-dimensional vectors and leverage similarity search mechanisms in vector spaces, offering significantly higher efficiency in knowledge retrieval and utilization compared to traditional databases [26].”
(2013) · Cited by 1,117
Retrieval-Augmented Generation for Large Language Models: A Survey
Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., Dai, Y., Sun, J., Wang, M., & Wang, H. (2023) · arXiv · doi:10.48550/arxiv.2312.10997 · Cited by 876 · cited 1×
“To better apply large language models to specific domains, increasing research focuses on the integration of external knowledge with language models, such as content planning [22], document retrieval [23], and language generation [24].”
The Second Conversational Intelligence Challenge (ConvAI2)
Dinan, E., Logacheva, V., Malykh, V., Miller, A., Shuster, K., Urbanek, J., Kiela, D., Szlam, A., Serban, I., Lowe, R., et al. (2019) · arXiv · doi:10.48550/arxiv.1902.00098 · Cited by 355 · cited 1×
“ConvAI2 [39] focuses on dialogues regarding personal interests, including 17,878 training pairs, 1000 validation pairs, and 1015 test pairs.”
The Second Conversational Intelligence Challenge (ConvAI2)
Dinan, E., Logacheva, V., Malykh, V., Miller, A., Shuster, K., Urbanek, J., Kiela, D., Szlam, A., Serban, I., Lowe, R., et al. (2020) · The Springer Series on Challenges in Machine Learning · doi:10.1007/978-3-030-29135-8_7 · Cited by 348
Bai, J., Bai, S., Chu, Y., Cui, Z., Dang, K., Deng, X., Fan, Y., Ge, W., Han, Y., Huang, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2309.16609 · Cited by 332 · cited 1×
“Baselines : To verify the generalization ability of the proposed method, we applied the model to two different base models (LLaMA-7B and Qwen-7B [42]) and conducted comparative experiments.”
Interactive Narrative: An Intelligent Systems Approach
Riedl, M. O., & Bulitko, V. (2013) · AI Magazine · doi:10.1609/aimag.v34i1.2449 · Cited by 299 · cited 1×
“However, pre-written dialogues may diminish the presence of NPCs and reduce gameplay flexibility [14], and the cost of manual creation is also high and time-consuming.”
Mixed-initiative co-creativity
Yannakakis, G. N., Liapis, A., & Alexopoulos, C. (2014) · OAR@UM (University of Malta) · Cited by 299
A Complete Survey on Generative AI (AIGC): Is ChatGPT from GPT-4 to GPT-5 All You Need?Preprint
Zhang, C.-N., Zhang, C.-S., Zheng, S., Qiao, Y., Li, C.-H., Zhang, M.-C., Dam, S. K., Thwal, C. M., Tun, Y. L., Huy, L. L., et al. (2023) · arXiv.org · doi:10.48550/arxiv.2303.11717 · Cited by 228 · cited 1×
“Subsequently, various procedural dialogue generation methods have emerged [15,16], among which neural network-based dialogue generation methods are the most prominent.”
Koo, M., & Yang, S.-W. (2025) · Encyclopedia · doi:10.3390/encyclopedia5010018 · Cited by 211 · cited 1×
“Therefore, this paper requires gpt-4o to score the generated sentences on hallucination mitigation, memory correctness, and character personalization, and calculates the average score (7-point Likert scale [40]) to represent the effectiveness of the model’s personalized generation capabilities.”
Character-LLM: A Trainable Agent for Role-Playing
Shao, Y., Li, L., Dai, J., & Qiu, X. (2023) · Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2023.emnlp-main.814 · Cited by 140
Character-LLM: A Trainable Agent for Role-Playing
Shao, Y., Li, L., Dai, J., & Qiu, X. (2023) · arXiv · doi:10.48550/arxiv.2310.10158 · Cited by 140 · cited 3×
“Character-LLMs [19] construct specific character experiences and materials, training more stable character agents through SFT technology.”
InfluentialHogan, A., Gutiérrez, C., Cochez, M., de Melo, G., Kirrane, S., Polleres, A., Roberto Navigli, A., Ngomo, A.-C. N., Rashid, S. M., Schmelzeisen, L., et al. (2022) · Synthesis lectures on data, semantics and knowledge · doi:10.1007/978-3-031-01918-0 · Cited by 117
Setting the scene: playing digital director in interactive storytelling and creation
Spierling, U., Grasbon, D., Braun, N., & Iurgel, I. (2002) · Computers & Graphics · doi:10.1016/s0097-8493(01)00176-5 · Cited by 107 · cited 1×
“Traditional NPC systems, which utilize pre-designed scripted rules or dialogue responses [1,2], often struggle to adapt content in real-time and fail to handle complex and varied player inputs, significantly limiting the authenticity of NPC interactions and the immersive experience of players.”
KG-BART: knowledge graph-augmented BART for generative commonsense reasoning
Liu, Y., Wan, Y., He, L., Peng, H., & Yu, P. S. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i7.16796 · Cited by 102
Bevilacqua, M., Blloshmi, R., & Navigli, R. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i14.17489 · Cited by 92 · cited 1×
“Given an input text T , we use SpringAMR [32], a BART-based pre-trained AMR parser [33], to encode the source sentence into an AMR graph G AMR ( T ) .”
Ashby, T., Webb, B. K., Knapp, G., Searle, J., & Fulda, N. (2023) · Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581441 · Cited by 63 · cited 1×
“Ashby et al. [30] manually crafted knowledge graphs of key game knowledge and employed large language models to create game dialogues.”
Fine-tuning GPT-2 on annotated RPG quests for NPC dialogue generation
van Stegeren, J., & Myśliwiec, J. (2021) · The 16th International Conference on the Foundations of Digital Games (FDG) 2021 · doi:10.1145/3472538.3472595 · Cited by 53 · cited 1×
“Stegeren et al. [20] fine-tuned GPT-2 to learn the structure of quest titles, objectives, and dialogues in “World of Warcraft”.”
Improving Personality Consistency in Conversation by Persona Extending
Liu, Y., Wei, W., Liu, J., Mao, X.-L., Fang, R., & Chen, D. (2022) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3511808.3557359 · Cited by 48 · cited 3×
“These include instruction-tuned LLaMA 7B-based models such as Alpaca 7B [43] and Vicuna 7B [44] and knowledge graph-enhanced generation models such as KG–BART [45] and DCRAG [46], as well as models optimized for character consistency [47].”
InfluentialDiversifying Content Generation for Commonsense Reasoning with Mixture of Knowledge Graph Experts
Yu, W., Zhu, C., Qin, L., Zhang, Z., Zhao, T., & Jiang, M. (2022) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2022.dlg4nlp-1.1 · Cited by 41
Diversifying Content Generation for Commonsense Reasoning with Mixture of Knowledge Graph ExpertsPreprint
Yu, W., Zhu, C.-G., Qin, L., Zhang, Z., Zhao, T., & Jiang, M. (2022) · Findings · doi:10.48550/arxiv.2203.07285 · Cited by 39 · cited 1×
“To better apply large language models to specific domains, increasing research focuses on the integration of external knowledge with language models, such as content planning [22], document retrieval [23], and language generation [24].”
Enhancing Dialogue Generation via Dynamic Graph Knowledge Aggregation
Tang, C., Zhang, H., Loakman, T., Lin, C., & Guérin, F. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-long.253 · Cited by 38
Graph Transformer for Graph-to-Sequence LearningPreprint
Cai, D., & Lam, W. P. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.07470 · Cited by 36 · cited 1×
“In the relation encoding stage, we refer to Cai’s [36] graph relation path reasoning method, employing a recurrent neural network with Gated Recurrent Units (GRUs) to encode the relationship between two concepts in the G –KG graph as a distributed representation.”
A Comprehensive Survey on Vector Database: Storage and Retrieval Technique, Challenge
Ma, L., Zhang, R., Han, Y., Yu, S., Wang, Z., Ning, Z., Zhang, J., Xu, P., Li, P., Qiao, Z., et al. (2023) · arXiv · doi:10.48550/arxiv.2310.11703 · Cited by 34 · cited 1×
“Vector databases store data [25] in the form of high-dimensional vectors and leverage similarity search mechanisms in vector spaces, offering significantly higher efficiency in knowledge retrieval and utilization compared to traditional databases [26].”
I Know What You Asked: Graph Path Learning using AMR for Commonsense Reasoning
Lim, J., Oh, D., Jang, Y., Yang, K., & Lim, H. (2020) · International Conference on Computational Linguistics · doi:10.18653/v1/2020.coling-main.222 · Cited by 33
A Survey on Game Playing Agents and Large Models: Methods, Applications, and ChallengesPreprint
Xu, X., Wang, Y.-X., Xu, C., Ding, Z.-L., Jiang, J., Ding, Z.-M., & Karlsson, B. F. (2024) · arXiv.org · doi:10.48550/arxiv.2403.10249 · Cited by 33 · cited 1×
“…the dialogue generation systems for NPCs are undergoing a paradigm shift from “scripted responses” to “cognitive intelligence”, with large-scale language models based on deep neural networks being widely applied to character personalization tasks [3], markedly enhancing NPC dialogue capabilities.”
A Step Towards the Future of Role-Playing Games: The SpyFeet Mobile RPG Project
Reed, A. A., Samuel, B., Sullivan, A., Grant, R., Grow, A., Lazaro, J., Mahal, J., Kurniawan, S., Walker, M. A., & Wardrip–Fruin, N. (2011) · Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · doi:10.1609/aiide.v7i1.12446 · Cited by 24 · cited 2×
“Traditional NPC systems, which utilize pre-designed scripted rules or dialogue responses [1,2], often struggle to adapt content in real-time and fail to handle complex and varied player inputs, significantly limiting the authenticity of NPC interactions and the immersive experience of players.”
Yu, J., Wu, S., Chen, J., & Zhou, W. (2024) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2024.findings-emnlp.794 · Cited by 12 · cited 1×
“These include instruction-tuned LLaMA 7B-based models such as Alpaca 7B [43] and Vicuna 7B [44] and knowledge graph-enhanced generation models such as KG–BART [45] and DCRAG [46], as well as models optimized for character consistency [47].”
NGEP: A Graph-based Event Planning Framework for Story Generation
Tang, C., Zhang, Z., Loakman, T., Lin, C., & Guérin, F. (2022) · Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the International Joint Conference on Natural Language Processing (AACL-IJCNLP) · doi:10.18653/v1/2022.aacl-short.24 · Cited by 9
What if Red Can Talk? Dynamic Dialogue Generation Using Large Language ModelsPreprint
Nananukul, N., & Wongkamjan, W. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2407.20382 · Cited by 9 · cited 2×
“Given that memory is a crucial factor in human personality generation [4], the current limitations in NPC memory mechanisms severely constrain the enhancement of their realism.”
NGEP: A Graph-based Event Planning Framework for Story GenerationPreprint
Tang, C., Zhang, Z., Loakman, T., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.10602 · Cited by 6 · cited 1×
“To better apply large language models to specific domains, increasing research focuses on the integration of external knowledge with language models, such as content planning [22], document retrieval [23], and language generation [24].”
Multi-Context Dependent Natural Text Generation for More Robust NPC Dialogue
Dai, K. (2020) · Digital Access to Scholarship at Harvard (DASH) (Harvard University) · Cited by 4
The NPC AI of The Last of Us: A case studyPreprint
Panwar, H. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2207.00682 · Cited by 3
Enhancing Dialogue Generation via Dynamic Graph Knowledge AggregationPreprint
Tang, C., Zhang, H., Loakman, T., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2306.16195 · Cited by 1
Yang, B., Zhao, K., Liu, D., et al. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2404.01129 · cited 1×
“AMR, as a widely used semantic representation method, has also been confirmed to improve the performance of LLMs as an intermediate representation [31].”
Temporal Knowledge-Graph Memory in a Partially Observable EnvironmentPreprint
Kim, T., François-Lavet, V., & Cochez, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2408.05861 · cited 1×
“Kim et al. [28] validated the effectiveness of knowledge graphs as memory systems, while Tang et al. [29] constructed multi-hop knowledge graphs to enhance dialogue generation.”
Graph Retrieval-Augmented Generation: A Survey
Peng, B., Zhu, Y., Liu, Y., Bo, X., Shi, H., Hong, C., Zhang, Y., & Tang, S. (2024) · arXiv · cited 1×
“Knowledge-enhanced dialogue generation, as a data-driven method, incorporates topological structures and knowledge, providing deeper semantic understanding capabilities for large language models [27].”
Research works printed but not linked: 3
- Bhutan, S., and Ali, N. (2025, April 24). AI in Gaming: Creating Realistic NPCs and Game Environments. Available online: https://www.researchgate.net/publication/377382632_AI_in_Gaming_Creating_Realistic_NPCs_and_Game_Environments.
- Kim, T., François-Lavet, V., and Cochez, M. (2024). Leveraging Knowledge Graph-Based Human-Like Memory Systems to Solve Partially Observable Markov Decision Processes. arXiv.
- Kingsbury, P., and Palmer, M. (2003, January 14–15). Propbank: The next level of treebank. Proceedings of the Treebanks and Lexical Theories, Citeseer, Växjö, Sweden.
Cited by
Rahman, A., Yu, A., & Cho, K. (2026) · Systems · doi:10.3390/systems14020175 · Cited by 2
Li, Y., Deng, J., Zhao, J., & Yim, J. (2026) · Applied Sciences · doi:10.3390/app16147254
Quest of Aivengarde: Comparative Study of Player Experience Across LLM Dialogue Systems
Rimer, E., Petersen, A. K. S., Ploug, R., & Scirea, M. (2026) · International Conference on the Foundations of Digital Games (FDG) · doi:10.1145/3815598.3815622
Metadata
- Authors
- Liu, X., Xie, Z., & Jiang, S.
- Affiliations
- Jiangnan University (CN)
- Wuxi Institute of Technology (CN)
- Corresponding author
- Xie, Zhenping
- Venue
- AI
- Publisher
- MDPI AG
- Published
- 2025-05-01
- Volume
- 6
- Issue
- 5
- Pages
- 93
- ISSN
- 2673-2688 (ISSN-L)
- OpenAlex
- W4410069631
- Cited by
- 3
- References
- 26
- Citations by year
- 2026: 3
- FWCI
- 1.21
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- AI (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Government of Jiangsu Province
- Qinglan Project of Jiangsu Province of China
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Artificial Intelligence in Games
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- non-player characters, personalized dialogue generation, role-playing games, character consistency, knowledge graphs, dynamic knowledge graphs, Abstract Meaning Representation, graph neural networks, memory retrieval, player immersion, large language models
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.