Short answers, pulled from the story.
The GloVe algorithm was created by researchers Jeffrey Pennington, Richard Socher, and Christopher Manning at Stanford University in 2014. They developed the method to translate human language into mathematical vectors using global statistics of word co-occurrence.
The core innovation of the GloVe algorithm is its use of global matrix factorization combined with local context window methods to create word vectors. This approach differs from word2vec by treating the entire text corpus as a single massive matrix of relationships rather than relying solely on local context windows.
The GloVe algorithm was launched in 2014 and utilized a corpus of six billion tokens to build its co-occurrence matrix. This massive dataset allowed the model to calculate probabilities of word pairs appearing together across the entire text library.
The GloVe algorithm encodes logical relationships as geometric offsets within the vector space to handle concepts like gender and royalty. For example, subtracting the vector for man from king and adding the vector for woman results in a vector close to queen.
The GloVe algorithm struggles with homographs because it calculates a single set of vectors for words with the same morphological structure. This limitation blurs the distinction between different meanings of words that share the same spelling.