Reproduce “Word Embeddings: A Survey” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Felipe Almeida, Geraldo Xexéo
This survey paper from the Federal University of Rio de Janeiro systematically describes the main strategies for building word embeddings, which are dense, fixed-length word vectors derived from co-occurrence statistics. It categorizes models into prediction-based and count-based approaches, detailing their evolution and highlighting their ability to encode linguistic regularities and their utility in various NLP tasks.
Key insights
- Despite diverse methodologies, all successful word embedding approaches are underpinned by the distributional hypothesis, inferring meaning from word co-occurrence statistics.
- The field is primarily categorized into two model types: prediction-based (e.g., Word2Vec) which learn from local contexts, and count-based (e.g., GloVe) which leverage global corpus statistics.
- Word embeddings encode linguistic regularities, capturing syntactic and semantic relationships as vector arithmetic, and their development was significantly driven by efforts to improve training efficiency and scalability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.