Beyond Monolithic LLMs: Modular AI for Online Harassment Detection
Bibliographic record
Abstract
Developing automated systems to analyze online harassment reports is a critical real-world challenge. In Canada, nearly one in five women experience online harassment annually, often resulting in users blocking others or deleting their accounts to protect themselves. We evaluate classic CNN-RNNs and modern SLMs/LLMs on SafeCity to address this challenge. Our findings show that different models excel at different harassment types: fine-tuned BERT (BERT-FT) performs best for “commenting,” CNN-RNN outperforms on “ogling,” and finetuned DeepSeek (DeepSeek-FT) achieves the highest accuracy for “groping.” To harness these complementary strengths, we introduce AD-ASH, a modular, specialist-driven system that assigns each sub-task to the most effective model. Beyond achieving improved performance on SafeCity dataset (0.66 exact-match ratio and 0.85 Hamming score), AD-ASH enables systematic comparative analysis across models and harassment types, offering valuable feedback for optimizing LLM performance through both architectural tuning and data-driven insights. This architecture also enabled a focused examination of dataset label suitability, which in turn motivated the future direction of integrating knowledge-graph-based contextual modeling to enhance RAG for LLMs. Overall, AD-ASH provides a robust, extensible framework for building hybrid AI systems tailored to the complexity and heterogeneity of real-world classification tasks. AD-ASH is a practical template for modular moderation systems where label heterogeneity undermines single-model performance.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".