design alignment taxonomies

This work addresses the challenge of efficiently constructing a comprehensive and well-structured taxonomy of artificial intelligence skills and tasks from massive hiring data. To this end, the authors propose TaxonomyBuilder, a framework that integrates systematic data filtering, clustering algorithms, and large language model–enhanced hierarchical label generation to automatically derive domain-specific taxonomies from curated, high-quality data subsets. Experimental results demonstrate that taxonomies built from filtered data exhibit significantly broader coverage and superior structural coherence compared to those generated from raw, unfiltered data using existing methods. The study thus establishes a novel paradigm for data-driven, automated taxonomy construction in specialized domains.

Current large language model alignment research overemphasizes static, universal values—helpfulness, harmlessness, and honesty—while neglecting competence adaptation, response timeliness, and audience heterogeneity. Method: We propose the first formally defined three-dimensional alignment framework—Competence, Transience, and Audience—grounded in conceptual modeling and context-sensitive analysis to establish a scalable, scenario-driven theory of dynamic alignment. We systematically map mainstream techniques—including RLHF, Constitutional AI, and context distillation—onto this 3D space to identify coverage gaps. Contribution/Results: The framework yields a precise, application-oriented alignment taxonomy, enabling fine-grained alignment design per use case. It shifts the paradigm from static value consistency toward dynamic functional applicability, advancing both theoretical foundations and practical deployment of aligned AI systems.

Automatic schema matching (ASM) suffers from low matching quality and excessive human intervention due to inherent complexity and uncertainty. Method: This study pioneers modeling ASM as a complex adaptive system (CAS) and introduces an agent-based modeling and simulation (ABMS) approach to construct a system-level matching framework exhibiting emergence and synergy. Departing from conventional local-rule-driven paradigms, the framework employs biologically inspired design and systems thinking to enable a paradigm shift—from atomic, isolated matching to global, self-organized coordination. Contribution/Results: The prototype tool Reflex-SMAS, built upon this framework, demonstrates significant improvements in matching accuracy and robustness across diverse scenarios. Empirical evaluation shows a reduction of over 62% in manual verification effort, confirming the dual advantages of the systemic approach: enhanced performance and substantial labor-cost savings.

Large language models (LLMs) deployed in organizational settings suffer from alignment failures—such as goal misgeneralization and discriminatory outputs—due to their black-box nature and information asymmetry during adoption; existing research lacks a systematic integration of organizational adoption processes with AI alignment mechanisms. Method: Drawing on principal-agent contract theory, this paper introduces the first theoretically grounded, end-to-end alignment framework for organizational LLM deployment, spanning “identification–evaluation–deployment–governance” stages. Through conceptual literature analysis, it develops the LLM ATLAS framework, which formally classifies stage-specific contractual governance mechanisms and alignment strategies. Contribution/Results: LLM ATLAS bridges the cognitive gap between organizational practitioners and LLM agents, establishing a rigorous theoretical foundation and actionable pathway for accountable, auditable, organization-level LLM governance. It advances both alignment science and organizational AI adoption theory by unifying technical alignment with institutional contract design.

Existing research treats entity set expansion, taxonomy expansion, and seed-guided taxonomy construction as disjoint tasks, resulting in poor method generalizability and a lack of unified modeling. Method: This paper proposes the first unified framework for all three taxonomy-related tasks, centered on collaborative learning of two structured reasoning skills—“finding sibling nodes” and “finding parent nodes”—enabled by taxonomy-guided instruction tuning, joint pretraining of both skills, and structure-aware prompting to foster skill complementarity and enhancement. Contribution/Results: The framework uncovers the shared skill foundation underlying diverse taxonomy tasks for the first time. It achieves state-of-the-art performance across all three tasks on multiple benchmarks, significantly improving generalizability and cross-task consistency while enabling unified modeling without task-specific architectural modifications.

This study addresses the high cost and expert dependency of manual taxonomy construction in software engineering (SE) by conducting the first systematic, multi-dimensional empirical evaluation of large language model (LLM)-driven automatic classification in this domain. Leveraging two representative approaches—TnT-LLM and CLIMB—and five state-of-the-art LLMs across seven human-annotated SE paper datasets, the work analyzes performance along key dimensions including classification quality, alignment with expert judgments, reliability, and efficiency. Results reveal that TnT-LLM achieves near-human classification quality but incurs high computational cost and structural complexity, whereas CLIMB offers 15–40× faster inference and 8–49× lower cost at the expense of reduced accuracy in tasks requiring deep technical reasoning. The findings elucidate critical trade-offs among quality, cost, and complexity, providing actionable guidance for method selection in practice.

Current AI alignment approaches struggle to manage conflicts and coordination among legitimate yet divergent values in pluralistic social contexts, largely due to a lack of understanding of how social values are organized and interact. This work addresses this gap by integrating sociological theories—such as role theory and field theory—into AI design, proposing a socially embedded, coordinative alignment paradigm. The approach employs role-based representations to model diverse perspectives and incorporates mechanisms for role activation, structured deliberation trajectories, and context-sensitive feedback loops to enable dynamic and accountable value coordination. By constructing a design space that supports structured, multi-perspective participation, this research lays the foundation for developing intelligent agents capable of effective, evaluable coordination in real-world social settings.

Current agent evaluation practices often reduce failures to system-level outcomes, making it difficult to pinpoint root causes or guide effective remediation. This work proposes an interaction-centric failure taxonomy and introduces, for the first time, a cross-architectural and generalizable framework for failure localization. The framework maps 41 distinct failure modes onto interaction edges between components—such as models, toolchains, and environments—and explicitly delineates responsibility boundaries among them. By integrating component interaction graph attribution, multi-source trajectory analysis, and an independent reasoning agent-based evaluator, the approach enables reproducible validation. Experiments across four state-of-the-art models demonstrate that the strongest evaluator achieves a Cohen’s κ of 0.76 with human annotations, confirming the taxonomy’s generalizability and consensus alignment.