1st Place - Asahi Newspaper NLP Competition 2023

Asahi NLP Competition Overview Image

*This is a conceptual illustration generated by AI. It may not accurately represent the actual competition or dataset.

1st Place - Asahi Newspaper NLP Competition 2023

Designed predictive models to identify viral news articles on social media using multimodal tabular data (text, images, metadata). Emsemble 3 models below.

  • Fine-tuned RoBERTa, which was pretrained by Asahi Newspaper team, for text embeddings. By combining meta information and texts data, I created new text data such as news.
  • deep / shallow LightGBM on engineered features:
    • Text (e.g., sentiment, headline length, word extraction)
    • Social metadata (shares, comments, etc.)
  • Achieved top F1 score: 78.39 out of 16 participants

Tools: Python, RoBERTa (Transformers), LightGBM

Official Site for details