Enhancing Machine Learning in Abusive Language Detection with Dataset Integration
Bibliographic record
Abstract
Abusive language detection models are widely reported to suffer from poor generalization, limiting their realworld effectiveness. This is largely due to sampling and lexical biases in datasets. In response to these issues, we aim to enhance the generalizability of abusive language detection models by leveraging and unifying existing datasets. We harmonize ten publicly available datasets under a consistent definition of abusive language and integrate them into a single dataset. Our core hypothesis is that while individual datasets exhibit sampling bias, their complementary characteristics can be harnessed to create a broader and more representative training distribution. To evaluate this hypothesis, we first empirically demonstrate the extent of sampling bias across datasets, then systematically integrate multiple datasets into an aggregated corpus and compare the classification performance of models trained on each individual dataset versus a model trained on the aggregated corpus using a held-out, uniformly sampled benchmark comprising data from all datasets. While the integrated model improves macro-F1 from 0.60 (average across single datasets) to 0.84. Furthermore, we quantify the contribution of each dataset to the integrated model's performance gains and its lexical dissimilarity relative to others, and find a strong correlation with a magnitude of 0.71. These findings suggest that integrating lexically diverse datasets exposes models to a broader spectrum of abuse-related language, mitigating dataset-specific sampling biases and enhancing model generalizability in real-world scenarios.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.014 | 0.038 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.002 | 0.005 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
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 source (direct Gemma or distilled Codex), 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".