Good, Better, Best: Choosing Word Embedding Context

TL;DR

This paper introduces two novel methods for learning word and phrase embeddings that integrate sentence context with dependency tree features, enhancing supervised term-matching performance.

Contribution

It presents new combined embedding techniques that leverage both contextual and structural information, improving upon existing methods.

Findings

Enhanced term-matching accuracy with combined embeddings

Neural network classifiers benefit from the proposed features

Structural and contextual features together outperform individual methods

Abstract

We propose two methods of learning vector representations of words and phrases that each combine sentence context with structural features extracted from dependency trees. Using several variations of neural network classifier, we show that these combined methods lead to improved performance when used as input features for supervised term-matching.