LoriensLibrary/cama-continuity-burden

--- license: cc-by-4.0 task_categories: - text-classification language: - en tags: - human-ai-interaction - memory-architecture - continuity-burden - affective-computing - longitudinal-case-study - CAMA pretty_name: "CAMA Continuity Burden Dataset" size_categories: - 1K<n<10K --- # CAMA Continuity Burden Dataset ## Overview This dataset provides aggregate research outputs from the **Circular Associative Memory Architecture (CAMA)** research program — a four-paper series investigating emotionally-indexed persistent memory for human-AI interaction. The central finding is the introduction and preliminary quantification of **continuity burden**: the communicative effort humans expend re-establishing context when interacting with memoryless AI systems. ## Key Findings - **Continuity reference rate**: 4.8 per 1,000 messages (320 references across 66,380 messages in 825 conversations) - **Scaling relationship**: Continuity references significantly increase with conversation length (R² = 0.381, p < 0.001) - 2.1% of conversations under 10 messages contain continuity references - 59.3% of conversations over 200 messages contain continuity references - **Category distribution**: Frustration markers dominate (56.6%), followed by context restoration (18.8%), re-explanation (15.9%), and temporal references (8.8%) - **Temporal decline**: Continuity reference rate peaked at 17.7 per 1,000 in February 2025, declining to 2.7–2.9 by late 2025, consistent with user adaptation - **CAMA coverage**: 52,602 emotionally annotated memories with 99.98% affect coverage ## Dataset Contents | File | Description | |------|-------------| | `paper4_stats.json` | Complete continuity burden analysis: reference counts, category breakdown, monthly rates, regression results, sample references | | `emergent_retention.json` | Analysis of AI behavioral retention indicators across 23,733 assistant messages | | `phase1_summary.json` | Descriptive statistics of the CAMA memory database | | `phase2_findings.json` | Comparative analysis between CAMA and pre-CAMA conditions | | `phase3_findings.json` | Architectural gap analysis including temporal coverage and inference confirmation | ## Important Limitations - **Single-participant longitudinal case study (n=1)** — findings cannot be generalized - Participant is the system designer, creating confounds - Continuity reference detection uses keyword matching without inter-rater reliability validation - Pre-CAMA and post-CAMA conditions overlap temporally; the distinction is architectural - All findings are exploratory and require multi-participant replication ## Research Papers This dataset supports the following published preprints: | Paper | Title | DOI | |-------|-------|-----| | 1 | Circular Associative Memory Architecture | [10.5281/zenodo.19051834](https://doi.org/10.5281/zenodo.19051834) | | 2 | Implementing Emotionally-Keyed Memory Retrieval in LLM Interfaces | [10.5281/zenodo.19052129](https://doi.org/10.5281/zenodo.19052129) | | 3 | CAMA Implementation and Functional Evaluation | [10.5281/zenodo.19192984](https://doi.org/10.5281/zenodo.19192984) | | 4 | Continuity Burden in Longitudinal Human-AI Interaction | [10.5281/zenodo.19226509](https://doi.org/10.5281/zenodo.19226509) | ## Source Code The CAMA architecture is open source: [github.com/LoriensLibrary/cama](https://github.com/LoriensLibrary/cama) ## Author **Angela Reinhold** - Lorien's Library LLC | Full Sail University - ORCID: [0009-0005-5803-8401](https://orcid.org/0009-0005-5803-8401) ## Privacy Notice This dataset contains **aggregate statistics only**. No personal conversation data, message content, or identifiable information is included. Raw interaction data is excluded to protect participant privacy. ## Citation ```bibtex @misc{reinhold2026continuityburden, author = {Reinhold, Angela}, title = {Continuity Burden in Longitudinal Human-AI Interaction: An Empirical Case Study of Emotionally-Indexed Persistent Memory}, year = {2026}, publisher = {Zenodo}, doi = {10.5281/zenodo.19226509} } ```许可证：CC BY 4.0 任务类别： - 文本分类 语言： - 英语 标签： - 人类-AI交互 - 记忆架构 - 连续性负担（continuity burden） - 情感计算 - 纵向案例研究 - CAMA 数据集名称："CAMA连续性负担数据集" 数据量范围：1K<n<10K # CAMA连续性负担数据集 ## 概述 本数据集收录了**循环关联记忆架构（Circular Associative Memory Architecture, CAMA）**研究项目的汇总研究成果——该项目由四篇系列论文组成，聚焦于面向人类-AI交互的情感索引式持久记忆研究。其核心研究发现为**连续性负担（continuity burden）**的提出与初步量化：即人类在与无记忆AI系统交互时，为重新建立上下文所付出的沟通成本。 ## 核心研究发现 - **连续性参考率**：每1000条消息出现4.8次（825场对话共66380条消息，累计320次连续性参考） - **缩放关系**：连续性参考次数随对话长度显著增加（决定系数R²=0.381，p<0.001） - 消息数少于10的对话中，仅2.1%包含连续性参考 - 消息数超过200的对话中，该比例升至59.3% - **类别分布**：挫折类标记占比最高（56.6%），其次为上下文恢复（18.8%）、重新解释（15.9%）与时间参照（8.8%） - **时间衰减趋势**：连续性参考率在2025年2月达到峰值——每1000条消息17.7次，至2025年末降至2.7~2.9次，与用户适应过程相符 - **CAMA覆盖范围**：共52602条带情感标注的记忆，情感覆盖度达99.98% ## 数据集内容 | 文件 | 描述 | |------|------| | `paper4_stats.json` | 完整的连续性负担分析报告：包含参考次数统计、类别拆分、月度发生率、回归分析结果与样本参考实例 | | `emergent_retention.json` | 针对23733条助手回复的AI行为保留指标分析 | | `phase1_summary.json` | CAMA记忆数据库的描述性统计结果 | | `phase2_findings.json` | CAMA与CAMA前阶段的对比分析 | | `phase3_findings.json` | 包含时间覆盖范围与推理验证的架构缺口分析 | ## 重要局限性 - **单参与者纵向案例研究（样本量n=1）**：研究结果无法进行外推推广 - 参与者为系统设计者，存在混淆变量 - 连续性参考检测采用关键词匹配方法，未进行评分者信度验证 - CAMA前后阶段在时间轴上存在重叠，二者的区分仅基于架构差异 - 所有研究发现均为探索性结论，需开展多参与者重复实验验证 ## 关联研究论文 本数据集支持以下已发表的预印本论文： | 序号 | 论文标题 | DOI | |------|----------|-----| | 1 | 循环关联记忆架构 | [10.5281/zenodo.19051834](https://doi.org/10.5281/zenodo.19051834) | | 2 | 在大语言模型（LLM）界面中实现情感键合记忆检索 | [10.5281/zenodo.19052129](https://doi.org/10.5281/zenodo.19052129) | | 3 | CAMA的实现与功能评估 | [10.5281/zenodo.19192984](https://doi.org/10.5281/zenodo.19192984) | | 4 | 纵向人类-AI交互中的连续性负担 | [10.5281/zenodo.19226509](https://doi.org/10.5281/zenodo.19226509) | ## 源代码 CAMA架构为开源项目：[github.com/LoriensLibrary/cama](https://github.com/LoriensLibrary/cama) ## 作者 **安吉拉·莱因霍尔德（Angela Reinhold）** - 洛里恩图书馆有限责任公司 | 福尔赛大学 - ORCID：[0009-0005-5803-8401](https://orcid.org/0009-0005-5803-8401) ## 隐私声明 本数据集仅包含汇总统计数据，未收录任何个人对话数据、消息内容或可识别信息。为保护参与者隐私，原始交互数据已被排除。 ## 引用 bibtex @misc{reinhold2026continuityburden, author = {Reinhold, Angela}, title = {Continuity Burden in Longitudinal Human-AI Interaction: An Empirical Case Study of Emotionally-Indexed Persistent Memory}, year = {2026}, publisher = {Zenodo}, doi = {10.5281/zenodo.19226509} }