Iteratively Linking Words Using Word2Vec and Cosine Similarity

We propose an algorithm that constructs relationships among any number of seed words. A relationship consists of a set of iteratively-generated paths of similar words, where each path links one seed word to another. The similar words are generated using Word2Vec word embeddings and the cosine similarity measure. By examining the effectiveness of the proposed algorithm in the mental health domain, we find that the algorithm effectively returns meaningful relationships and has the potential to be used for hypothesis generation and information extraction.