Group-Sparse Matrix Factorization for Transfer Learning of Word Embeddings

The paper introduces a two-stage estimator for transfer learning of word embeddings using group-sparse matrix factorization, addressing the challenge of adapting embeddings to new domains with limited data. It combines large-scale corpora with domain-specific data, proving that it can achieve high accuracy with less domain-specific input by altering only a few embeddings. The approach demonstrates efficient computation and establishes the first bounds on group-sparse matrix factorization, offering a potentially significant advancement for practitioners in natural language processing seeking to improve domain adaptation.