Salmon Run: Clustering Word Vectors using a Self Organizing Map

Friday, October 03, 2014 Clustering Word Vectors using a Self Organizing Map Continuing on from last week's experiments with Neural Networks (NN), I use the same dataset of 97k sentences to visualize latent relationships between the words in these sentences. To do this, I first trained a Word2Vec NN with word 4-grams from this sentence corpus, and then used the transition matrix to generate word vectors for each of the words in the vocabulary. Using the word vectors, I trained a Self Organizing Map (SOM) , another type of NN, which allowed me to locate each word on a 50x50 grid.

Read full article from Salmon Run: Clustering Word Vectors using a Self Organizing Map