Identifying and Mitigating Gender Bias in Hyperbolic Word Embeddings
This addresses gender bias in natural language processing for users of hyperbolic embeddings, but it is incremental as it extends existing bias studies to a new embedding type.
The paper tackled gender bias in hyperbolic word embeddings by proposing a new measure, gyrocosine bias, and a debiasing method called Poincaré Gender Debias (PGD), which effectively reduced bias with minimal semantic offset in experiments.
Euclidean word embedding models such as GloVe and Word2Vec have been shown to reflect human-like gender biases. In this paper, we extend the study of gender bias to the recently popularized hyperbolic word embeddings. We propose gyrocosine bias, a novel measure for quantifying gender bias in hyperbolic word representations and observe a significant presence of gender bias. To address this problem, we propose Poincaré Gender Debias (PGD), a novel debiasing procedure for hyperbolic word representations. Experiments on a suit of evaluation tests show that PGD effectively reduces bias while adding a minimal semantic offset.