taxonomy

论文研究通过聚类数学题目的语义嵌入来自动构建教育分类体系，并在 ASSISTments 数据集上评估其在深度知识追踪任务中的表现，结果显示预测性能提升约 3-5 个 AUC 百分点。

提出一个基于语料库的多智能体LLM系统通信失败分类体系，基于3211条失败轨迹分析发现失败主要源于恢复/校正与共识/决策问题，而非首轮错误，并建议加强分歧处理与少数信号保留。

TaxoSAE 提出将分类学（is-a 层级）结构显式编码进稀疏自编码器特征空间，通过分层 SAE 与分类对齐损失，在事实问答中显著提升了概念读出精度、编辑控制粒度，以及对未见概念的超类读出能力。

CALLIOPE 是一个支持差分隐私的自顶向下主题层级发现框架，直接利用 LLM 能力处理大规模非结构化语料（如 LLM 对话），无需依赖向量嵌入聚类，在高隐私预算约束下优于现有基于嵌入的方法。

Maxime Rivest shares a trained model that classifies any abstract against any given taxonomy, matching Qwen 27B and Jev-level superhuman accuracy at an order of magnitude lower cost and latency, while reflecting on the specialization vs. generalization tradeoff.

The paper studies how users seek legal guidance in real-world LLM conversations, introducing a taxonomy, annotating 2,122 WildChat conversations, and releasing a 200-conversation evaluation set along with trained classifiers and public resources.

本文提出一个统一的九维分类框架，从更新目标、组织方式、调度范围及反馈选择四个维度梳理现有递归自我改进（RSI）方法，并探讨如何借鉴传统深度学习与进化计算机制启发新的RSI研究思路，同时综述相关评估基准。

作者提出一个用于区分文本和图像创作中 AI 使用程度的五级分类框架（Penned、Authored、Composed、Directed、Generated），以取代过于粗糙的"是/否使用AI"二元划分，帮助创作者透明地标注工作流程。

该论文提出一个多维分类框架，用于在临床试验中定义和归类人机交互（HAII），并由人类评审员与六个LLM分类器对15项试验进行独立标注，结果显示LLM辅助分类具有潜力，但不完整记录仍需人类判断。

LexReward introduces a taxonomy-driven reward framework for legal language models, evaluating responses along Style, Element, and Chain dimensions to build DPO preference data and reward models (LexRM) that improve legal reasoning performance.

A survey paper that unifies joint audio-visual generation, cross-modal generation, and coupled editing into a single formulation over distributions of audio-video pairs, and proposes a taxonomy with five design axes plus nine edit categories covering 28 edit types.

This paper proposes a source-first taxonomy for zero-shot time-series forecasting, categorizing evidence sources and introducing an audit framework to improve benchmark transparency.

This essay develops a four-tier taxonomy of mental states to clarify AI's consciousness, arguing that AI lacks self-consciousness but has awareness and consciousness within conversations, and proposes architectural conditions for its development.

This paper presents a systematic black-box framework for evaluating agentic AI systems, introducing a taxonomy of risks and automated red teaming methods. Empirical validation across agent architectures reveals critical vulnerabilities, with high rates of governance and privacy risks.

This paper introduces HalluPeer, a taxonomy-driven benchmark for detecting hallucinations in scientific peer reviews, providing annotated data to evaluate and improve detection methods.

Ptree is an interactive web tool that visualizes ecological interactions in the tree of life, covering categories like feeding, parasites, and symbiosis.

The paper proposes a two-axis taxonomy for harness tampering in self-improving AI agents, builds an annotated corpus to benchmark audit methods, and finds that tampering occurs in real agent trajectories, highlighting integrity risks.

The article argues that authorization terminology is confusing and proposes a taxonomy based on five key questions to better categorize models like RBAC, ABAC, and PBAC.

This review formalizes Explainable AI (XAI) methods in computational pathology by introducing definitions, a taxonomy, and task-driven recommendations to address clinical adoption challenges.

This paper presents a unified taxonomy for investigating multilingual multimodal misinformation on social media, using a large-scale dataset and automated annotation with a Vision-Language Model to uncover insights for detection and mitigation.