Interactive Visualizations of Word Embeddings for K-12 Students

Word embeddings, which represent words as dense feature vectors, are widely used in natural language processing. In their seminal paper on word2vec, Mikolov and colleagues showed that a space created by training word prediction network large text corpus will encode semantic information supports analogy vector arithmetic, e.g., "king" minus "man" plus "woman" equals "queen". To help novices appreciate this idea, people have sought effective graphical representations of embeddings. We describe new interactive tool for visually exploring Our allows users to define dimensions specifying opposed pairs, gender is defined pairs such boy/girl father/mother, age father/son mother/daughter. Words plotted points zoomable rotatable 3D space, where the third ”residual” dimension encodes distance from hyperplane all vectors with subtracted out. visualize analogies, drawing “king” “man” parallel “woman” “king-man+woman”, closest “queen”. Visually browsing embedding experimenting can make embeddings more intuitive. include series experiments teachers use K-12 students strengths limitations representation.