Domain-Specific Word Embeddings with Structure Prediction

TL;DR

This paper introduces W2VPred, a novel word embedding method that simultaneously captures general, domain-specific, and structural information, enabling dynamic and aligned embeddings across different corpora and domains.

Contribution

The paper presents a new embedding approach that models structure between sub-corpora and domains, outperforming baselines in analogy and structure prediction tasks.

Findings

W2VPred outperforms baselines in analogy tests.

It effectively predicts structure without prior information.

Demonstrated usefulness in Digital Humanities research.

Abstract

Complementary to finding good general word embeddings, an important question for representation learning is to find dynamic word embeddings, e.g., across time or domain. Current methods do not offer a way to use or predict information on structure between sub-corpora, time or domain and dynamic embeddings can only be compared after post-alignment. We propose novel word embedding methods that provide general word representations for the whole corpus, domain-specific representations for each sub-corpus, sub-corpus structure, and embedding alignment simultaneously. We present an empirical evaluation on New York Times articles and two English Wikipedia datasets with articles on science and philosophy. Our method, called Word2Vec with Structure Prediction (W2VPred), provides better performance than baselines in terms of the general analogy tests, domain-specific analogy tests, and multiple specific word embedding evaluations as well as structure prediction performance when no structure is given a priori. As a use case in the field of Digital Humanities we demonstrate how to raise novel research questions for high literature from the German Text Archive.