Towards News Aggregation in Russian: a BERT-based Approach to News Article Similarity Detection
Anna Valer'evna Glazkova · Computational Linguistics and Intellectual Technologies · 2021
This paper presents neural models developed for the Dialogue Evaluation 2021: Russian News Clustering shared task.The participants were offered the news clustering task, but solutions based on the classification were also accepted.We applied a classification approach and implemented a hard-voting ensemble of pretrained language models fine-tuned on the official competition dataset.Our solution ranked third among 17 teams at the Russian News Clustering shared task in both private and public leaderboards and obtained 95.73% of F1-score (private LB) and 96.5% (public LB).