Persona-Grounded Dialogue: Advancing Conversational AI Through Personality and Context Integration

Persona-grounded dialogue represents a fundamental paradigm shift in conversational AI, moving beyond generic responses to create dialogue systems that maintain consistent personalities and character traits throughout extended interactions. This approach addresses one of the most persistent challenges in dialogue systems: the inconsistency problem where models generate contradictory responses for similar inputs, undermining user trust and engagement. By grounding responses in well-defined persona information—encompassing background facts, personality traits, linguistic behaviors, and interaction styles—these systems achieve remarkable improvements in consistency, engagement, and human-like interaction quality[1][2][3]. Recent advances have demonstrated that persona-grounded approaches can achieve up to 200% improvements in BLEU scores and 247% improvements in ROUGE scores compared to traditional dialogue systems[4][5].

Understanding Persona in Conversational Context

Defining Persona Components

The concept of persona in dialogue systems encompasses multiple interconnected dimensions that collectively define a conversational agent's identity[1][2][6]. Background facts and profiles form the foundational layer, providing concrete information about demographics, occupation, interests, and personal history that grounds the agent's responses in specific contexts[7][8]. Personality traits constitute the psychological dimension, often modeled using established frameworks such as the Big Five personality model (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism), which influence emotional expression and conversational style[1][9][10].

Linguistic behavior represents the surface manifestation of personality through word choice, sentence structure, formality levels, and discourse patterns[1][8]. Research has demonstrated that different personality traits correlate with distinct linguistic signatures: extraverted users tend to employ more enthusiastic language and social expressions, while neurotic individuals may exhibit hesitancy and worry-focused vocabulary[11][10]. Interaction style encompasses broader conversational patterns, including turn-taking preferences, topic initiation tendencies, and response length preferences[1][9].

Evolution from Static to Dynamic Personas

Traditional approaches relied heavily on static persona descriptions—fixed sets of sentences describing character attributes that remained constant throughout conversations[7][8]. The seminal PersonaChat dataset exemplified this approach, providing crowd-sourced persona descriptions such as "I enjoy skiing" or "I have blonde hair" that dialogue agents could reference during response generation[7][12]. However, static personas suffer from several limitations: they fail to capture the natural evolution of personality expression over time, struggle with implicit personality manifestations, and lack the flexibility to adapt to conversational dynamics[13][4][14].

Contemporary research has shifted toward dynamic persona modeling, where persona information is continuously extracted, updated, and refined based on ongoing dialogue interactions[4][13][15]. This paradigm recognizes that human personality expression is contextual and evolving, requiring systems that can adapt their understanding of user personas through accumulated conversational experience[13][16][14].

Technical Approaches and Architectures

Memory Network-Based Systems

Memory networks emerged as one of the earliest successful approaches for integrating persona information into dialogue generation[2][3][8]. The Ranking Profile Memory Network utilizes ranking loss functions to score candidate responses against persona consistency, while the Key-Value Profile Memory Network employs sophisticated retrieval mechanisms to identify relevant persona information for each conversational turn[1][2]. These systems maintain explicit memory stores containing persona facts and utilize attention mechanisms to selectively activate relevant memories during response generation[8][17].

The effectiveness of memory network approaches lies in their ability to perform targeted retrieval of persona-relevant information without overwhelming the generation process with irrelevant details[2][6]. However, these systems face scalability challenges when dealing with large persona databases and struggle with implicit personality manifestations that aren't explicitly stored in memory[13][8].

Transformer-Based Integration

The introduction of transformer architectures revolutionized persona-grounded dialogue by enabling more sophisticated context understanding and persona integration[18][19][20][8]. BERT-based approaches have demonstrated exceptional performance in persona consistency tasks, achieving accuracy rates exceeding 90% through bidirectional context encoding that captures subtle persona-context relationships[18][20][21].

Advanced transformer implementations employ multi-head attention mechanisms specifically designed for persona-grounded tasks. These systems utilize separate attention heads for persona encoding, dialogue history processing, and response generation, enabling fine-grained control over how personality information influences output generation[19][8][17]. The integration of persona information occurs at multiple levels: input embeddings incorporate persona representations, attention layers weight persona relevance dynamically, and output layers ensure consistency with established personality traits[19][8].

Large Language Model Approaches

The emergence of large language models has introduced new possibilities for persona-grounded dialogue, particularly through prompt engineering and in-context learning[22][23][12][24]. GPT-based systems can maintain persona consistency through carefully crafted prompts that embed personality descriptions directly into the generation context[24][10]. These approaches benefit from the extensive pre-training of LLMs, which provides robust understanding of personality-language relationships without requiring task-specific training data[22][24].

However, LLM-based approaches face unique challenges in persona control and consistency[25][26][10]. Research has revealed that LLMs can sometimes generate responses that contradict established personas, particularly in extended conversations where context windows become constrained[25][10]. Post-Persona Alignment (PPA) frameworks address these limitations by implementing two-stage generation processes: initial response generation followed by persona-guided refinement[13]. This approach allows models to generate natural responses while ensuring final outputs align with established personality traits[13].

Dataset Landscape and Benchmarks

Foundational Datasets

The PersonaChat dataset remains the cornerstone of persona-grounded dialogue research, comprising 162,064 utterances between crowdworkers assigned specific personas[7][23]. This dataset established the fundamental paradigm of persona-grounded dialogue and continues to serve as a primary evaluation benchmark[2][3][27]. The dataset's strength lies in its authentic human-human conversations where speakers genuinely attempt to embody assigned personalities, providing realistic examples of persona-consistent dialogue[7][8].

ConvAI2 extended PersonaChat into a competitive evaluation framework, facilitating direct comparison between different persona-grounded approaches[3][23]. The Multi-Session Chat (MSC) dataset addressed limitations in single-session approaches by providing multi-session dialogues that require long-term persona consistency[27][8]. These datasets have enabled systematic evaluation of persona consistency across extended interactions, revealing significant challenges in maintaining personality coherence over time[13][27].

Contemporary Developments

Recent dataset development has focused on addressing the limitations of early benchmarks through increased scale, diversity, and sophisticated persona modeling[28][23][12]. ComperDial provides comprehensive evaluation metrics for persona consistency, including fine-grained assessments of personality trait adherence and response quality[28][29]. The dataset incorporates multiple scored responses for each dialogue turn, enabling more robust evaluation of automatic dialogue metrics[28].

PersonalityChat represents a significant advancement by integrating Big Five personality traits with persona descriptions, enabling research into personality-driven dialogue generation[23][11]. This dataset demonstrates how established psychological frameworks can enhance persona modeling, providing more nuanced and theoretically grounded personality representations[23][10]. PSYDIAL focuses specifically on personality-based synthetic dialogue generation, utilizing LLMs to create datasets that capture personality nuances across different cultural contexts[10][30].

The shift toward synthetic dataset generation using LLMs has enabled unprecedented scale in persona dialogue data[12][24][31]. Google's Synthetic-Persona-Chat dataset comprises 20,000 conversations generated using a Generator-Critic architecture framework, demonstrating that high-quality persona dialogues can be created synthetically while maintaining natural conversation patterns[12][31].

Evaluation Methodologies and Metrics

Consistency Measurement

Persona consistency evaluation requires sophisticated metrics that can assess both explicit fact adherence and implicit personality manifestation[28][32][33]. The C-Score provides overall consistency assessment by evaluating response alignment with established persona facts[13][34]. However, simple fact-checking approaches fail to capture the nuanced ways personality influences language use, necessitating more sophisticated evaluation frameworks[28][32].

Natural Language Inference (NLI) based evaluation has emerged as a powerful approach for assessing persona consistency[35][36][34]. These methods utilize trained NLI models to determine whether generated responses logically follow from established persona information, providing more nuanced consistency assessment than simple keyword matching[35][15][37]. The Q² metric demonstrates this approach by using automatic question generation and question answering to evaluate factual consistency in knowledge-grounded dialogue[35][36].

Quality and Engagement Assessment

Beyond consistency, persona-grounded dialogue systems must balance personality adherence with response quality and user engagement[1][27][38]. Research has revealed tension between these objectives: highly persona-consistent responses may become repetitive or constrained, while engaging responses may sacrifice personality coherence[1][38]. The Persona-F1 metric addresses this challenge by measuring the overlap between generated content and established persona facts while accounting for response naturalness[27][38].

Human evaluation studies remain essential for assessing the subjective qualities of persona-grounded dialogue[28][12][27]. These evaluations typically assess multiple dimensions: fluency (linguistic quality), consistency (persona adherence), coherence (logical flow), engagingness (user interest), and humanness (natural interaction quality)[28][27]. Recent research has demonstrated significant variation in human perception of personality consistency, highlighting the need for comprehensive evaluation frameworks that account for individual differences in personality perception[28][39].

Current Challenges and Limitations

Cross-Domain Generalization

One of the most persistent challenges in persona-grounded dialogue involves cross-domain generalization[15][37]. Models trained on datasets like PersonaChat, which primarily feature "real-world" scenarios, struggle to maintain persona consistency when deployed in different narrative contexts such as fantasy settings or specialized domains[15][37]. This limitation significantly constrains the practical applicability of persona-grounded systems, as real-world deployments often involve domain shifts that weren't present in training data[15].

The domain adaptation problem is particularly acute for persona extraction models, which must identify personality-relevant information from dialogue in new contexts[15][37]. Research has demonstrated that persona extraction accuracy can degrade substantially when models encounter unfamiliar vocabulary, cultural references, or interaction patterns[37][40]. Natural Language Inference approaches have shown promise in addressing these challenges by providing post-hoc adaptation mechanisms that don't require extensive retraining[15][37].

Long-Term Consistency Maintenance

Multi-session dialogue consistency represents another significant challenge, as persona-grounded systems must maintain personality coherence across extended interactions spanning multiple conversation sessions[13][27][8]. Traditional approaches that concatenate dialogue history face computational constraints as conversations extend, leading to context truncation that can compromise persona consistency[13][27]. Memory management becomes critical as systems must balance retaining persona-relevant information with computational efficiency[13][8].

Persona drift emerges as conversations extend, where subtle inconsistencies accumulate over time, gradually eroding the coherence of established personality traits[13][34]. This phenomenon is particularly problematic in real-world applications where users may engage with systems across multiple sessions over extended periods[13][8]. Post-Persona Alignment frameworks address these challenges by implementing explicit consistency checking and refinement stages, though computational overhead remains a concern[13].

Evaluation Complexity

The multifaceted nature of persona evaluation creates significant assessment challenges[28][27][33]. Simple automated metrics may fail to capture subtle personality manifestations, while human evaluation is expensive and subject to individual variation in personality perception[28][39][33]. The development of reliable automatic evaluation metrics for persona consistency remains an active research area, with recent work exploring LLM-based evaluation approaches[28][41].

Cultural and linguistic diversity in personality expression adds another layer of evaluation complexity[42][10][30]. Persona-grounded systems must account for cultural differences in personality manifestation, communication styles, and social interaction norms[10][30]. The limited availability of persona dialogue datasets in languages other than English constrains research into cross-cultural persona modeling[42][27][10].

Emerging Technologies and Future Directions

Advanced Integration Frameworks

Recent research has introduced sophisticated frameworks that integrate multiple aspects of persona modeling into unified systems[13][8][43]. Multi-attribute control systems enable fine-grained manipulation of different personality dimensions, allowing developers to create more nuanced and controllable persona-grounded agents[43][44]. These systems utilize latent space modeling to disentangle different personality aspects, enabling independent control of traits like language style, emotional expression, and social interaction patterns[43].

Retrieval-Augmented Generation (RAG) approaches for persona-grounded dialogue represent another promising direction[45][46][16]. These systems maintain dynamic persona knowledge bases that can be updated and expanded based on ongoing interactions, enabling more flexible and adaptive personality modeling[16][14]. The integration of knowledge graphs with persona information provides structured representations that support more sophisticated reasoning about personality-context relationships[45][8].

Synthetic Data Generation and Augmentation

The application of Large Language Models for synthetic persona dialogue generation has opened new possibilities for dataset creation and augmentation[12][24][10]. Advanced generation pipelines can create diverse persona dialogues at scale, addressing the data scarcity issues that have constrained persona-grounded dialogue research[12][24][10]. Generator-Critic architectures ensure quality control in synthetic data generation, with specialized models evaluating generated conversations for persona consistency and dialogue quality[12].

Persona augmentation techniques address dataset bias issues by generating alternative personality expressions for existing personas[47][8][38]. These approaches recognize that single persona descriptions may not capture the full range of personality expression, necessitating data augmentation strategies that explore different manifestations of the same underlying personality traits[47][38].

Real-World Application Integration

Industry deployment of persona-grounded dialogue systems has revealed practical considerations that extend beyond research benchmarks[48][49][50]. Commercial applications require robust persona management systems that can handle user privacy concerns, personality adaptation over time, and integration with existing customer service workflows[48][49]. The development of cloud-based persona services enables smaller organizations to leverage sophisticated persona-grounded dialogue capabilities without requiring extensive in-house expertise[48].

Therapeutic and educational applications represent emerging domains for persona-grounded dialogue systems[50][9]. Mental health applications require careful consideration of personality adaptation to support therapeutic goals, while educational systems must balance engaging personality expression with pedagogical effectiveness[50][51]. These applications highlight the broader societal implications of persona-grounded dialogue technology.

Conclusion

Persona-grounded dialogue represents a fundamental advancement in conversational AI, transforming generic chatbots into sophisticated agents capable of maintaining consistent personalities and adapting to user preferences through extended interactions. The field has evolved from simple static persona descriptions to dynamic, multi-dimensional personality modeling that captures the complexity of human conversational behavior. Current state-of-the-art systems demonstrate remarkable capabilities in persona consistency and user engagement, with transformer-based approaches and large language models achieving unprecedented performance levels.

The integration of established psychological frameworks, particularly the Big Five personality model, has provided theoretical grounding that enhances both system performance and evaluation validity. Advanced architectures employing memory networks, attention mechanisms, and multi-stage generation processes enable fine-grained control over personality expression while maintaining response quality and naturalness. The development of sophisticated evaluation metrics has moved beyond simple consistency checking to encompass the multifaceted nature of personality assessment.

Despite significant progress, substantial challenges remain in cross-domain generalization, long-term consistency maintenance, and robust evaluation methodologies. The tension between personality consistency and response diversity continues to require careful balancing, while cultural and linguistic variations in personality expression demand more inclusive research approaches. The computational complexity of advanced persona-grounded systems presents ongoing challenges for real-world deployment, particularly in resource-constrained environments.

Future developments in persona-grounded dialogue will likely focus on more sophisticated integration of multimodal information, enhanced cross-cultural personality modeling, and improved adaptation mechanisms that can learn and evolve personas through ongoing interactions. The emergence of synthetic data generation techniques promises to address dataset scarcity issues, while advances in evaluation methodologies will enable more reliable assessment of system performance across diverse application domains.

The broader implications of persona-grounded dialogue extend beyond technical achievements to encompass fundamental questions about human-computer interaction, user privacy, and the role of artificial personalities in society. As these systems become more sophisticated and widely deployed, careful consideration of ethical implications and user welfare will become increasingly important. The field's continued evolution will require interdisciplinary collaboration between computer scientists, psychologists, linguists, and ethicists to ensure that advances in persona-grounded dialogue technology benefit users and society broadly.

The convergence of large language models, sophisticated persona modeling, and robust evaluation frameworks positions persona-grounded dialogue as a cornerstone technology for next-generation conversational AI. Success in this domain will require continued innovation in technical approaches, expanded research into diverse cultural contexts, and thoughtful consideration of the broader societal implications of increasingly sophisticated artificial personalities.

출처
[1][Conversational AI 13강] Grounding in Persona - Sungyeon Kim https://goldstaryeon.tistory.com/401
[2] Persona-Based Conversational AI: State of the Art and ... https://arxiv.org/pdf/2212.03699.pdf
[3] Persona-Based Conversational AI: State of the Art and ... https://arxiv.org/abs/2212.03699
[4] Towards Personalized Dialogue Without Pre-defined ... https://aclanthology.org/2024.emnlp-main.581/
[5] Towards Personalized Dialogue Without Pre-defined ... https://arxiv.org/abs/2403.03102
[6][Literature Review] Dialogue Language Model with Large ... https://www.themoonlight.io/en/review/dialogue-language-model-with-large-scale-persona-data-engineering
[7] arXiv:1801.07243v5 [cs.AI] 25 Sep 2018 https://arxiv.org/pdf/1801.07243.pdf
[8] Dialogue Language Model with Large-Scale Persona Data ... https://aclanthology.org/2025.naacl-industry.71.pdf
[9] The Role of Personality in Conversational AI https://peritushub.com/blog/the-role-of-personality-in-conversational-ai/
[10] Personality-based Synthetic Dialogue Generation using ... https://arxiv.org/html/2404.00930v1
[11] Exploring the Impact of Personality Traits on ... https://arxiv.org/html/2504.12313v1
[12] Faithful Persona-based Conversational Dataset ... https://aclanthology.org/2024.nlp4convai-1.8/
[13] Post Persona Alignment for Multi-Session Dialogue ... https://arxiv.org/html/2506.11857v1
[14] Generating, retrieving persona and ... https://peerj.com/articles/cs-2979.pdf
[15] Using Natural Language Inference to Improve Persona ... https://arxiv.org/html/2401.06742v1
[16] Doing Personal LAPS: LLM-Augmented Dialogue ... https://dl.acm.org/doi/pdf/10.1145/3626772.3657815
[17] Dialogue Language Model with Large-Scale Persona Data ... https://aclanthology.org/2025.naacl-industry.71/
[18] Performance Analysis of Transformer Based Models (BERT ... https://arxiv.org/abs/2308.04950
[19] Enhancing Fake News Detection with Transformer Models ... https://etasr.com/index.php/ETASR/article/view/10678
[20] Fake News Detection: Comparative Evaluation of BERT ... https://arxiv.org/html/2412.14276v2
[21] Fake News Detection: Comparative Evaluation of BERT ... https://arxiv.org/html/2412.14276v1
[22] Explainable Fake News Detection With Large Language ... https://arxiv.org/abs/2405.03371
[23] PersonalityChat: Conversation Distillation for Personalized ... https://arxiv.org/html/2401.07363v1
[24] Generating Personalized Dialogues using GPT-3 https://aclanthology.org/2022.ccgpk-1.4/
[25] Can LLM-Generated Misinformation Be Detected? https://arxiv.org/html/2309.13788v5
[26] Fake News Generation, Explanation and Detection of ... https://arxiv.org/html/2310.05046v2
[27] Recent Trends in Personalized Dialogue Generation https://arxiv.org/html/2405.17974v1
[28] ComperDial: Commonsense Persona-grounded Dialogue ... https://arxiv.org/pdf/2406.11228.pdf
[29] ComperDial: Commonsense Persona-grounded Dialogue ... https://arxiv.org/abs/2406.11228
[30][논문 리뷰] PSYDIAL: Personality-based Synthetic ... https://www.themoonlight.io/ko/review/psydial-personality-based-synthetic-dialogue-generation-using-large-language-models
[31] google-research-datasets/Synthetic-Persona-Chat https://github.com/google-research-datasets/Synthetic-Persona-Chat
[32] Evaluating Task-Oriented Dialogue Consistency through ... https://arxiv.org/html/2407.11857v1
[33] A Comprehensive Assessment of Dialog Evaluation Metrics https://aclanthology.org/2021.eancs-1.3.pdf
[34] Building Persona Consistent Dialogue Agents with Offline ... https://aclanthology.org/2023.emnlp-main.110.pdf
[35] Evaluating Factual Consistency in Knowledge-Grounded ... https://aclanthology.org/2021.emnlp-main.619/
[36] $Q^{2}$: Evaluating Factual Consistency in Knowledge ... https://arxiv.org/abs/2104.08202
[37] Using Natural Language Inference to Improve Persona ... https://arxiv.org/abs/2401.06742
[38] Improving Persona Consistency in Dialogue Generation ... https://arxiv.org/html/2508.06886v1
[39] Trustworthiness in Conversational Agents: Patterns in User ... https://thesai.org/Downloads/Volume15No12/Paper_3-Trustworthiness_in_Conversational_Agents.pdf
[40] Dialogue act-based partner persona extraction for ... https://www.sciencedirect.com/science/article/abs/pii/S0957417424012466
[41] PairEval: Open-domain Dialogue Evaluation Metric with ... https://openreview.net/forum?id=y6aGT625Lk
[42] nu-dialogue/real-persona-chat · Datasets at ... https://huggingface.co/datasets/nu-dialogue/real-persona-chat
[43] Towards Personalized Dialogue Generation with Latent ... https://openreview.net/forum?id=eCXfUq3RDf
[44] Personality-affected Emotion Generation in Dialog Systems https://dl.acm.org/doi/10.1145/3655616
[45] Towards Robust Fact-Checking: A Multi-Agent System with ... https://arxiv.org/abs/2506.17878
[46] Towards explainable fake news detection and automated ... https://www.sciencedirect.com/science/article/pii/S0925231224012219
[47] Optimizing training data for persona-grounded dialogue via ... https://yonsei.elsevierpure.com/en/publications/optimizing-training-data-for-persona-grounded-dialogue-via-synthe
[48] ‎Conversational AI Persona Application | Sprinklr Help Center https://www.sprinklr.com/help/categories/conversational-ai-persona-application/63e3b808c070bb1342399ae3
[49] How Conversational AI Humanizes Chatbot Personality https://www.coveo.com/blog/chatbot-personality/
[50] Building Personality-Adaptive Conversational AI for Mental ... https://dl.acm.org/doi/10.1145/3698587.3701489
[51] The role of personality traits in predicting educational use ... https://www.nature.com/articles/s41598-025-16339-0
[52] Learning to Predict Persona Information for Dialogue ... https://www.microsoft.com/en-us/research/publication/learning-to-predict-persona-information-for-dialogue-personalization-without-explicit-persona-description/
[53] Faithful Persona-based Conversational Dataset ... https://openreview.net/forum?id=tSqMro2ZUP
[54] AIcrowd | Commonsense Persona-Grounded Dialogue ... https://www.aicrowd.com/challenges/commonsense-persona-grounded-dialogue-challenge-2025
[55] Persona-Based Conversational AI: State of the Art and ... https://www.computer.org/csdl/proceedings-article/icdmw/2022/460900a993/1KBr4jldVW8
[56] Enhancing Engagement in Persona-Grounded Dialogue ... https://www.dbpia.co.kr/journal/articleDetail?nodeId=NODE11488258
[57][PDF] Persona-Based Conversational AI: State of the Art ... https://www.semanticscholar.org/paper/a054e491539f76cc2da4bb7a823c4195b7d90535
[58] Optimizing training data for persona-grounded dialogue via ... https://www.sciencedirect.com/science/article/abs/pii/S0957417424026630
[59] Personalized Dialogue Agents with Dynamically Adapted ... https://arxiv.org/html/2406.13960v1
[60] Recent Trends in Personalized Dialogue Generation - LREC http://www.lrec-conf.org/proceedings/lrec-coling-2024/pdf/2024.main-1.1192.pdf
[61] A Compact Knowledge-grounded Persona-based ... https://dl.acm.org/doi/10.1145/3711857
[62] A Persona-Grounded Chit-Chat Generation Framework https://aclanthology.org/2020.nlp4convai-1.14.pdf
[63] 페르소나 기반의 장기 대화를 위한 다각적 어텐션을 활용한 ... https://koreascience.kr/article/CFKO202306643316448.pdf
[64] Personalized Topic Transition for Dialogue System https://aclanthology.org/2024.yrrsds-1.5.pdf
[65] What affects the usage of artificial conversational agents ... https://www.sciencedirect.com/science/article/abs/pii/S0747563223001395
[66] 24 Ways to Ground Yourself When You're Feeling ... https://laconciergepsychologist.com/blog/24-ways-to-ground-yourself-when-youre-feeling-overwhelmed/
[67] Grounding Techniques: 30 Examples & How They Help https://www.choosingtherapy.com/grounding-techniques/
[68] GROUNDING https://oklahomatfcbt.org/wp-content/uploads/2018/08/Grounding-Worksheet.pdf
[69] 13 Grounding Techniques To Help Calm Anxiety https://health.clevelandclinic.org/grounding-techniques
[70] Dialogue Language Model with Large-Scale Persona Data ... https://arxiv.org/abs/2412.09034
[71] Grounding Techniques | Article https://www.therapistaid.com/therapy-article/grounding-techniques-article
[72] Dialogue response coherency evaluation with feature ... https://www.sciencedirect.com/science/article/abs/pii/S0952197625006098
[73] 10+ Mindful Grounding Techniques (Incl. Group Exercise) https://positivepsychology.com/grounding-techniques/
[74] Debiasing large language models for persona-based ... https://openreview.net/forum?id=mA3eY3E7nk
[75] Automatic Evaluation of Consistency in Dialogue System ... https://www.cl.ecei.tohoku.ac.jp/publications/2024/dthesis_shiki_sato.pdf
[76] 5 Grounding Techniques for Overstimulated HSPs https://highlysensitiverefuge.com/5-grounding-techniques/
[77] Personality-Driven Negotiation Dialogue Generation https://www.emergentmind.com/topics/personality-driven-argumentation-based-negotiation-dialogue-generation-pan-dg
[78] Improving Personality Consistency in Conversation by ... https://dl.acm.org/doi/10.1145/3511808.3557359
[79] Learning to improve persona consistency in conversation ... https://www.sciencedirect.com/science/article/abs/pii/S0950705121005086
[80] Beyond Discrete Personas: Personality Modeling Through ... https://aclanthology.org/2025.coling-main.470.pdf
[81] PAED: Zero-Shot Persona Attribute Extraction in Dialogues https://aclanthology.org/2023.acl-long.544.pdf
[82] Persona-Based Conversational AI: State of the Art and ... https://www.alphaxiv.org/overview/2212.03699v1
[83] Extracting Personalities from past Conversations? - API https://community.openai.com/t/extracting-personalities-from-past-conversations/33551
[84] Improving Personality Consistency in Conversation by ... https://openreview.net/pdf?id=90KaW3XjgRq