Content Meeting Session – 3rd September 2024

🌐 Discussion Topic:
A Comparison of Correspondence Analysis with PMI-Based Word Embedding Methods

🔍 Overview of the Presentation
This session explored the integration of Correspondence Analysis (CA) with popular PMI-based word embedding techniques like GloVe and Word2Vec. The research presented a novel link between CA and PMI, demonstrating how square-root and root-root transformations can significantly improve the performance of CA for NLP tasks, particularly word similarity evaluations. The paper also highlighted how these transformations help mitigate extreme values in word-context matrices.

🚨 Key Insights:

• Enhanced performance of CA using square-root and root-root transformations
• Theoretical relationships between CA and PMI-based methods
• Improved word embeddings through dimensionality reduction

🌟 Presenter:

This insightful session was led by Qianqian Qi, sharing valuable insights from her research. Peter van der Heijden, a co-author of the paper, also took part in the meeting, providing his expertise and enriching the discussion.

For further information, you can review the full paper here.