Interpretable Neural Embeddings with Sparse Self-Representation

Minxue Xia, Hao Zhu

Interpretability benefits the theoretical understanding of representations. Existing word embeddings are generally dense representations. Hence, the meaning of latent dimensions is…