Learning Effective and Interpretable Semantic Models using Non-Negative Sparse Embedding

In this paper, we introduce an application of matrix factorization to produce corpus-derived, distributional<br/>models of semantics that demonstrate cognitive plausibility. We find that word representations<br/>learned by Non-Negative Sparse Embedding (NNSE), a variant of matrix factorization, are sparse,<br/>effective, and highly interpretable. To the best of our knowledge, this is the first approach which<br/>yields semantic representation of words satisfying these three desirable properties. Though extensive<br/>experimental evaluations on multiple real-world tasks and datasets, we demonstrate the superiority<br/>of semantic models learned by NNSE over other state-of-the-art baselines.