Checklist ROAI 2025 Selection Camp – Test 2 · Task 1
How Toxic Are You Online?
Multi-label classification of online comments into toxic, severe_toxic, obscene and insult without transformer models.
The task
Online platforms need automatic detection of harmful comments. Each comment may carry any subset of four labels: toxic, severe_toxic, obscene and insult.
The contestant trains a classifier on the labelled comments and predicts the four binary labels for every test comment. Transformer models are not allowed.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train_data.csv(id,comment_text, toxic,severe_toxic, obscene, insult) andtest_data.csv.- You submit
- CSV with subtaskID (always 1), datapointID (comment id) and answer as a list [toxic,
severe_toxic, obscene, insult] of 0/1 values. - Scoring
- Per-label F1 mapped to points and summed (maximum 100): toxic F1 0.85–0.905 → 1–25 points;
severe_toxic0.10–0.350 → 0–25; obscene 0.65–0.750 → 0–25; insult 0.60–0.710 → 0–25; at or above the maximum F1 a label earns full points, at or below the minimum only the minimum, and in between the score is interpolated between the minimum and (maximum − 5). A separate evaluation is made on a 30% partial subset. - Rules
- Transformer models are not allowed
- On-site during the national training and selection camp (17–24 June 2025, POLITEHNICA Bucharest)
- Internet limited to the contest platform and the documentation of the available libraries
- Only submissions ticked during the round count; anti-fraud session locking on the judge
- Format
- 2025 national selection camp, test 2, 23 June 2025 (06:00–11:00 UTC on the judge).