Word embedding dimensionality reduction using dynamic variance thresholding (DyVaT)

Avraham Treistman, Dror Mughaz, Ariel Stulman, Amit Dvir

Abstract: Highlights•Variance thresholding to reduce dimensionality requires manual threshold selection.•Dynamic variance thresholding (DyVaT) uses the kneedle algorithm to set a threshold.•DyVat can be used recursively to create large ontologies using a single root word.