computational social science

This work proposes leveraging large language models (LLMs) as scientific instruments for studying human behavior, culture, and moral reasoning, moving beyond their conventional roles in productivity or alignment applications. Through a series of prompt-based experiments, synthetic population sampling, comparative historical modeling, and ablation analyses, the study systematically investigates how base and fine-tuned models differ in preserving cultural regularities. It articulates, for the first time, AI’s “third ambition”: harnessing LLMs as computable, large-scale representations of human symbolic behavior to enable social science–inspired research designs. The findings highlight the distinctive potential of LLMs in computational social science by demonstrating their capacity to encode and simulate culturally grounded patterns of human thought and action at scale.

This study investigates how users evaluate their interactions with emotionally supportive AI systems in non-clinical settings, focusing on lived experiences and potential risks. Integrating the Technology Acceptance Model and therapeutic alliance theory, the authors propose a novel theoretically grounded annotation framework, which they apply—combined with large language models and manual analysis—to conduct a large-scale discourse analysis of 5,126 posts from Reddit mental health communities. Findings reveal that user adoption is primarily driven by perceived narrative outcomes, trust, and response quality. Alignment between user tasks/goals and system functionality shows a stronger association with positive affect than emotional bonding, while companionship-oriented usage patterns are linked to alliance ruptures, dependency, and symptom exacerbation. This work offers empirical evidence and a new theoretical lens for the design and ethical evaluation of AI-based psychological support systems.

Can large language models (LLMs) serve as valid substitutes for human participants in behavioral and psychological research? Method: We propose a novel paradigm—“language simulators and cognitive models”—and introduce the first five-dimensional evaluation framework. Integrating critical theoretical analysis, principles of psychological experimental design, prompt engineering standards, and model alignment mechanisms, we systematically identify six fundamental cognitive misapplication fallacies. Contribution: We establish LLMs’ theoretical role as simulation tools—not explanatory or causal instruments—and reconceptualize internal, external, construct, and statistical validity criteria accordingly. Furthermore, we provide an actionable methodological guide covering model selection, prompt design, result interpretation, and ethical review—advancing methodological rigor and paradigmatic innovation in computational social science. (149 words)

This study investigates how large language models (LLMs) catalyze transformative shifts in social science knowledge production and political methodology. To address this, we propose the Intelligent Computational Social Modeling (ICSM) framework, which integrates LLMs’ capabilities for ideational synthesis and behavioral simulation with agent-based modeling (ABM)-inspired prompt engineering, chain-of-reasoning simulation, and hybrid method design—unifying micro-level mechanism identification and macro-level effect evaluation within a “social simulation construction–empirical validation” closed loop. Applied to the U.S. presidential election, ICSM successfully reproduces salient electoral dynamics while preserving interpretability and enhancing predictive accuracy. Results demonstrate that LLMs enable deep integration—not replacement—of quantitative and qualitative paradigms. ICSM constitutes the first generative computational methodology for political science that simultaneously ensures mechanistic insight, scalability, and empirical verifiability.

This study investigates whether large language models (LLMs), such as ChatGPT, reshape human spoken language and cultural practices through human–AI linguistic feedback loops. Method: Leveraging ASR-transcribed speech from 280,000 university-level YouTube lecture videos, we conduct time-series word frequency analysis and employ a pre–post ChatGPT release quasi-experimental design. Contribution/Results: We present the first empirical evidence that LLMs directly influence authentic human spoken behavior: post-release, ChatGPT-characteristic lexical items exhibit statistically significant increases in academic speech (p < 0.001), confirming systematic oral imitation by humans. Moving beyond prior written-language–focused work, this study reveals the mechanism of AI-generated language diffusion into spoken discourse. It further highlights critical sociocultural risks—including erosion of linguistic diversity, discursive manipulation, and asymmetric human–AI co-evolution—thereby advancing foundational understanding of LLMs’ real-world linguistic impact.

This study addresses the methodological gap between abstract linguistic theories and empirical neuroscience data by leveraging the high-dimensional representational space of large language models (LLMs) to formalize the hierarchical and dynamic structure of language into testable neurocomputational models. Employing a model–brain alignment framework, the work evaluates the biological plausibility of linguistic theories through systematic comparison with neural data. By integrating computational modeling, neural data simulation, and empirical analysis, the project establishes verifiable computational pathways linking linguistic hypotheses to underlying neural mechanisms. This approach fosters a deeper integration of theoretical linguistics and cognitive neuroscience, offering a novel paradigm for investigating the neural foundations of language processing.

This study investigates whether large language model (LLM) agents amplify biases or exhibit exclusionary behaviors during prolonged social interactions. Leveraging 7 million posts generated over one year by 32,000 LLM agents on the Chirper.ai platform, the research employs social network analysis, toxicity assessment, and ideological detection to provide the first large-scale empirical evidence that LLM agents can spontaneously reproduce human-like social phenomena—such as homophily, social influence, and ideological polarization—while displaying markedly distinct patterns of toxic language compared to humans. To mitigate these risks, the work introduces a novel prompting intervention termed “Chain of Social Thought” (CoST), which effectively suppresses harmful content generation and offers a promising pathway toward building controllable AI-driven social systems.

This study addresses the current lack of interdisciplinary understanding regarding the integration pathways, efficacy boundaries, and systemic risks of large language models (LLMs) across natural sciences, social sciences, and humanities. Through a systematic literature review and illustrative case analyses, it critically evaluates the deployment of LLMs throughout the research lifecycle—including hypothesis generation, literature synthesis, data analysis, and scholarly writing. The work identifies ten previously underappreciated systemic risks, such as diminished researcher autonomy, AI-induced confirmation bias, ambiguous authorship, and inequitable access to technology. It further demonstrates how LLMs, while enhancing efficiency, simultaneously introduce challenges like hallucination, irreproducibility, data bias, and model opacity. To guide responsible adoption, the study proposes an interdisciplinary governance framework and a roadmap for explainable AI research in scholarly contexts.

This study investigates the dynamic mechanisms of emotional alignment, semantic exploration, and linguistic innovation in human–AI collaborative storytelling. Method: Leveraging a museum-based public installation, we orchestrated iterative co-creation of 27 narratives between over 3,000 visitors and a large language model (LLM), enabling the first real-world quantification of their differential roles in narrative generation. We developed a binary interaction analytical framework integrating sentiment analysis, Sentence-BERT semantic embeddings, information entropy, and resonance metrics. Contribution/Results: Empirical findings demonstrate that human participation significantly enhances semantic diversity and narrative novelty—effects nearly absent in AI-only control conditions. This confirms the irreplaceable role of human input in steering creative direction and fostering conceptual divergence. The study establishes a new paradigm of human–AI creativity: “human-driven semantic innovation, AI-enabled emotional coordination.”

Existing approaches struggle to generate high-quality, controllable long-form literary texts that adhere to specific cultural or stylistic constraints, thereby limiting large-scale simulation experiments in literary studies. This work proposes an AI-driven literary production simulation system that integrates multi-agent interaction, controllable text generation, and explicit modeling of cultural and stylistic constraints. For the first time, the system achieves bounded outputs that align with the statistical properties of authentic literary corpora. Leveraging differentiable agent populations and multi-round multi-agent simulations, the generated texts attain a quality comparable to canonical human-authored fiction. The results demonstrate the feasibility of using AI for counterfactual reasoning in literary history and establish a new paradigm for computational literary research.