# Which Paper Is Newer? Data Description

*English translation by SOTA – AI Community of the Russian original. Licensed CC BY-SA 3.0, like the original. Organisers who would like this translation removed can email sota.ai.community@gmail.com.*

*Day 2 of the final of AI Olymp, the Republican Olympiad in Artificial Intelligence of Kazakhstan, 2025. This is the Data tab ("Dataset Description") of the Kaggle competition "Kazakhstan Respa Final - Day 2 (afterparty)"; unlike the Overview, it is in Russian only. Original: [kaggle.com/competitions/kazakhstan-respa-final-day-2-late-competition/data](https://www.kaggle.com/competitions/kazakhstan-respa-final-day-2-late-competition/data). The Data tab lists 4 files (98.5 MB): `train_timeseries.csv`, `train_sentences.csv`, `test_sentences.csv` and `sample_submission.csv`.*

**train_timeseries.csv** - the same dataset as in the homework: 140 categories and the number of papers on each topic published on arXiv for each day (from 2000 up to April 2025)

**train_sentences.csv** - the first 2 sentences of papers. For each of the 140 topics we sampled, for each category, at least 10 papers per year. The data are made slightly harder because each paper can have several categories.

**test_sentences.csv** - from all the papers not selected for training we sampled 400 papers for each topic, evenly over the whole history (uniformly at random). For each paper we extracted the first 2 sentences of the abstract. We then split the resulting pool at random into pairs, guaranteeing that within each pair there is a difference of at least 2 years. Your task is to determine, for each pair, which paper was newer. Output 0 if the first paper was newer, and 1 if the second paper was newer.