Analysis of Word Embeddings Using Fuzzy Clustering

Abstract

In data dominated systems and applications, a concept of representing words in a numerical format has gained a lot of attention. There are a few approaches used to generate such a representation. An interesting issue that should be considered is the ability of such representations - called embeddings - to imitate human-based semantic similarity between words. In this study, we perform a fuzzy-based analysis of vector representations of words, i.e., word embeddings. We use two popular fuzzy clustering algorithms on count-based word embeddings, known as GloVe, of different dimensionality. Words from WordSim-353, called the gold standard, are represented as vectors and clustered. The results indicate that fuzzy clustering algorithms are very sensitive to high-dimensional data, and parameter tuning can dramatically change their performance. We show that by adjusting the value of the fuzzifier parameter, fuzzy clustering can be successfully applied to vectors of high - up to one hundred - dimensions. Additionally, we illustrate that fuzzy clustering allows to provide interesting results regarding membership of words to different clusters.