2019/01/22 by Hwiyeol Jo, Jo, Hwiyeol, Ceyda Cinarel +1
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Text and Document Classification Technologies #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1901.07651
openalex publication_date 2019/01/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We propose a novel and simple method for semi-supervised text classification.\nThe method stems from the hypothesis that a classifier with pretrained word\nembeddings always outperforms the same classifier with randomly initialized\nword embeddings, as empirically observed in NLP tasks. Our method first builds\ntwo sets of classifiers as a form of model ensemble, and then initializes their\nword embeddings differently: one using random, the other using pretrained word\nembeddings. We focus on different predictions between the two classifiers on\nunlabeled data while following the self-training framework. We also use\nearly-stopping in meta-epoch to improve the performance of our method. Our\nmethod, Delta-training, outperforms the self-training and the co-training\nframework in 4 different text classification datasets, showing robustness\nagainst error accumulation.\n