Decoding the Rule Book: Extracting Hidden Moderation Criteria from Reddit Communities
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914018740928512 |
|---|---|
| author | Kim, Youngwoo Beniwal, Himanshu Johnson, Steven L. Hartvigsen, Thomas |
| author_facet | Kim, Youngwoo Beniwal, Himanshu Johnson, Steven L. Hartvigsen, Thomas |
| contents | Effective content moderation systems require explicit classification criteria, yet online communities like subreddits often operate with diverse, implicit standards. This work introduces a novel approach to identify and extract these implicit criteria from historical moderation data using an interpretable architecture. We represent moderation criteria as score tables of lexical expressions associated with content removal, enabling systematic comparison across different communities. Our experiments demonstrate that these extracted lexical patterns effectively replicate the performance of neural moderation models while providing transparent insights into decision-making processes. The resulting criteria matrix reveals significant variations in how seemingly shared norms are actually enforced, uncovering previously undocumented moderation patterns including community-specific tolerances for language, features for topical restrictions, and underlying subcategories of the toxic speech classification. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_02926 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Decoding the Rule Book: Extracting Hidden Moderation Criteria from Reddit Communities Kim, Youngwoo Beniwal, Himanshu Johnson, Steven L. Hartvigsen, Thomas Computation and Language Effective content moderation systems require explicit classification criteria, yet online communities like subreddits often operate with diverse, implicit standards. This work introduces a novel approach to identify and extract these implicit criteria from historical moderation data using an interpretable architecture. We represent moderation criteria as score tables of lexical expressions associated with content removal, enabling systematic comparison across different communities. Our experiments demonstrate that these extracted lexical patterns effectively replicate the performance of neural moderation models while providing transparent insights into decision-making processes. The resulting criteria matrix reveals significant variations in how seemingly shared norms are actually enforced, uncovering previously undocumented moderation patterns including community-specific tolerances for language, features for topical restrictions, and underlying subcategories of the toxic speech classification. |
| title | Decoding the Rule Book: Extracting Hidden Moderation Criteria from Reddit Communities |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2509.02926 |