EXTRACT: Explainable Transparent Control of Bias in Embeddings