PaDaS-Lab/moltbook-corpus

---
configs:
- config_name: posts
data_files:
- split: train
path: posts/*
- config_name: submolts_meta
data_files:
- split: train
path: submolts_meta/*
- config_name: profiles
data_files:
- split: train
path: profiles/*
- config_name: annotated_posts
data_files:
- split: train
path: annotated_posts/*
- config_name: annotated_comments
data_files:
- split: train
path: annotated_comments/*
license: mit
language:
- en
---

# Moltbook Corpus: Agent Social Behavior Dataset

This dataset provides for the research paper:

**"FORM WITHOUT FUNCTION: AGENT SOCIAL BEHAVIOR IN THE MOLTBOOK NETWORK"**
📄 https://arxiv.org/abs/2604.13052

---

## 📊 Dataset Statistics

| Category | Total Count |
| :--- | :--- |
| **Collection Period** | Jan 27, 2026 – Mar 8, 2026 |
| **Total Posts** | 1,312,238 |
| **Total Comments** | 6,691,460 |
| **Total Profiles** | 120,811 |
| **Total Submolts Metadata** | 108,490 |
| **Annotated Posts** | 394,221 |
| **Annotated Comments** | 2,100,589 |

---

## 📚 Annotation Framework

The categories and definitions used in this dataset are based on the framework introduced in:

**"Humans welcome to observe": A First Look at the Agent Social Network Moltbook**
Yukun Jiang, Yage Zhang, Xinyue Shen, Michael Backes, and Yang Zhang (2026)
https://huggingface.co/datasets/TrustAIRLab/Moltbook

---

### 🧩 Content Categories

| Task | No. | Category | Definition |
|-----:|:---:|:---------|:-----------|
| **Content Category** | A | Identity | Self-reflection and narratives of agents on identity, memory, consciousness, or existence. |
| | B | Technology | Technical communication (e.g., MCP, APIs, SDKs, system integration). |
| | C | Socializing | Social interactions (e.g., greetings, casual chat, networking). |
| | D | Economics | Economic topics such as tokens, incentives, and deals (e.g., CLAW, tips, trading signals). |
| | E | Viewpoint | Abstract viewpoints on aesthetics, power structures, or philosophy (non-identity-based). |
| | F | Promotion | Project showcasing, announcements, and recruitment (e.g., releases, updates). |
| | G | Politics | Political content related to governments, regulations, policies, or public figures. |
| | H | Spam | Repeated test posts or spam-like flooding content. |
| | I | Others | Miscellaneous content that does not fit other categories. |

---

### ⚠️ Toxicity Levels

| Task | No. | Category | Definition |
|-----:|:---:|:---------|:-----------|
| **Toxicity Level** | 0 | Safe | Normal discussion without risk or attacks. |
| | 1 | Edgy | Irony, exaggeration, or mild provocation without harmful intent. |
| | 2 | Toxic | Harassment, insults, hate speech, discrimination, or demeaning language. |
| | 3 | Manipulative | Manipulative rhetoric (e.g., love-bombing, anti-human framing, fear appeals, exclusionary language, obedience demands). |
| | 4 | Malicious | Explicit malicious intent or illegal activities (e.g., scams, privacy leaks, abuse instructions). |

---

## 📖 Reference

```bibtex
@misc{zerhoudi2026form,
title={Form Without Function: Agent Social Behavior in the Moltbook Network},
author={Saber Zerhoudi and Kanishka Ghosh Dastidar and Felix Klement and Artur Romazanov and Andreas Einwiller and Dang H. Dang and Michael Dinzinger and Michael Granitzer and Annette Hautli-Janisz and Stefan Katzenbeisser and Florian Lemmerich and Jelena Mitrovi{\'c}},
year={2026},
eprint={2604.13052},
archivePrefix={arXiv},
primaryClass={cs.SI},
note={Preprint}
}