Checklist ROAI 2026 Selection Camp Qualifier · Task 2
Fake friends..
Separate noise from news texts and cluster the news into an unknown number of categories.
The task
A database of categorised news texts was scrambled by the owner's friends: the category labels were lost and noise texts were inserted. Only the scrambled test file is available.
Subtask 1 asks whether each text is noise (1) or news (0). Subtask 2 asks for a category for every text (−1 for noise, 1…n for news), where neither the categories nor their number is known. The model BAAI/bge-small-en-v1.5 may be used.
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
test_data.csvcontaining the scrambled texts.- You submit
- CSV with subtaskID (1 or 2), datapointID (text id) and answer.
- Scoring
- Subtask 1 (40 points): MCC; 40 at MCC ≥ 0.87, linear from 4 to 40 for 0 ≤ MCC < 0.87, 4 below 0. Subtask 2 (60 points): adjusted mutual information; 60 at AMI ≥ 0.53, linear from 6 to 60 for 0 ≤ AMI < 0.53, 6 below 0.
- Rules
- Allowed embedding model:
BAAI/bge-small-en-v1.5 - On-site at Genesis College, Bucharest, through the Nitro Contestant Cloud
- Only the packages listed in the round rules may be used
- Only submissions selected by the contestant count towards the final ranking
- Allowed embedding model:
- Format
- ROAI 2026 qualifier for the selection camp, 8 April 2026.