This paper introduces a method for improving Random Indexing (RI) embeddings by averaging them on a sparse Positive Pointwise Mutual Information (PPMI) graph. The technique showed a significant accuracy increase from 19.4% to 30.7% on a fairytale corpus for semantic analogy questions related to Google family categories. However, the method did not outperform neural baselines on other datasets like text8 and SimLex-999, and even reduced accuracy for some other embedding types. AI
IMPACT This research offers a non-gradient method to enhance specific embedding types, though its broader competitiveness with neural models remains limited.
RANK_REASON The item is a research paper detailing a novel method for improving word embeddings. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- Bloom filter
- continuous bag-of-words model
- Positive Pointwise Mutual Information
- PPMI+SVD
- Random Indexing
- SimLex-999
- singular value decomposition
- Skip-gram
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
