Extracting keywords from email data using distributed word vectors

Current keyword extraction methods often use statistical models to select certain words from a set of documents, which fail to take advantage of the information available in the documents themselves. We propose a model for identifying semantic relationships between words in a document to identify keywords that more accurately capture the meaning of the document. Specifically, we use distributed word vectors to learn hypernymy and test the relationship on the contents of emails to deduce meaningful keywords. We find that this model is capable of producing a list of words that contain such keywords, though more analysis is necessary on learning the hypernymy relationship between words to quantitatively narrow this list to only those keywords.