Checklist AICC 2026 Round 7 · Task 1
Polarity
Classify word pairs as synonyms or antonyms from 50 labelled examples, using only bert-large-uncased.
The task
A water-damaged book of word pairs recorded whether each pair were synonyms (for example happy/glad) or antonyms (for example high/low). Only fifty pairs still show their label; several hundred have lost it.
The training set has 50 labelled pairs (25 synonyms, 25 antonyms). The test set has 686 unlabelled pairs; the data page states that half are synonyms and half antonyms. Every word is a single token in the model's vocabulary, and no training word appears in the test set.
Contestants predict 0 (synonyms) or 1 (antonyms) for each test pair, and may use bert-large-uncased in any way.
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.csv(50 rows: w1, w2, label) andtest.csv(686 rows:row_id, w1, w2).- You submit
submission.csvwith columnsrow_id,label, one row per test pair.- Scoring
- Macro-averaged F1 over the two classes, score = (F1(0) + F1(1)) / 2. Baseline 0.49, reference solution 0.91.
- Rules
- bert-large-uncased (loaded from Hugging Face with transformers) is the only pretrained model allowed; no external lexicon, thesaurus or word list (including nltk.corpus) and no external text corpus.
- Internet access only to download the model.
- Individual participation (maximum team size 1); at most 30 submissions per day.
- AICC contest rules (stated on each Kaggle rules page, not enforceable): no use of LLMs for writing code or getting task ideas; no internet use other than official library documentation and the contest platform; no communication with anyone during the contest; clarifications only via the #clarification-requests channel on the AICC Discord server.
- Format
- AICC Round 7, online on Kaggle, 22 May 2026 19:58 UTC – 24 May 2026 20:01 UTC.