Guided Semi-Supervised Non-Negative Matrix Factorization

Classification and topic modeling are popular techniques in machine learning that extract information from large-scale datasets. By incorporating a priori such as labels or important features, methods have been developed to perform classification tasks; however, most can both do not allow for guidance of the topics features. In this paper, we propose novel method, namely Guided Semi-Supervised Non-negative Matrix Factorization (GSSNMF), performs by supervision pre-assigned document class user-designed seed words. We test performance method on legal documents provided California Innocence Project 20 Newsgroups dataset. Our results show proposed improves accuracy coherence comparison past (SSNMF), (Guided NMF), Topic Supervised NMF.