Reverse Dictionary Using an Improved CBoW Model

Abstract

A reverse dictionary is a tool that solves the Tip-of-the-Tongue problem by predicting the target word given a meaning or description. This paper proposes a modified Continuous Bag-Of-Words(CBOW) model inspired by the Recurrent Neural Network(RNN), to implement a reverse dictionary. The model possesses the merits of both the CBOWmodel and RNN while stripping away the complexity associated with RNN. We evaluate the model by measuring accuracy based on the top-2 predictions.