X-Class: Text Classification with Extremely Weak Supervision

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3043–3053 June 6–11, 2021. ©2021 Association for Computational Linguistics 3043 X-Class: Text Classification with Extremely Weak Supervision Zihan Wang 1 Dheeraj Mekala 1 Jingbo Shang 1,2 1 Department of Computer Science and Engineering, University of California San Diego, CA, USA 2 Halıcıo˘glu Data Science Institute, University of California San Diego, CA, USA {ziw224, dmekala, jshang}@ucsd.edu Abstract In this paper, we explore text classification with extremely weak supervision, i.e., only relying on the surface text of class names. This is a more challenging setting than the seed-driven weak supervision, which allows a few seed words per class. We opt to attack this problem from a representation learning perspective—ideal document representations should lead to nearly the same results between clustering and the desired classification. In par- ticular, one can classify the same corpus differ- ently (e.g., based on topics and locations), so document representations should be adaptive to the given class names. We propose a novel framework X-Class to realize the adaptive rep- resentations. Specifically, we first estimate class representations by incrementally adding the most similar word to each class until in- consistency arises. Following a tailored mix- ture of class attention mechanisms, we obtain the document representation via a weighted av- erage of contextualized word representations. With the prior of each document assigned to its nearest class, we then cluster and align the doc- uments to classes. Finally, we pick the most confident documents from each cluster to train a text classifier. Extensive experiments demon- strate that X-Class can rival and even outper- form seed-driven weakly supervised methods on 7 benchmark datasets. 1 Introduction Weak supervision has been recently explored in text classification to save human effort. Typical forms of weak supervision include a few labeled docu- ments per class (Meng et al., 2018; Jo and Cinarel, 2019), a few seed words per class (Meng et al., 2018, 2020a; Mekala and Shang, 2020; Mekala et al., 2020), and other similar open-data (Yin et al., 2019).Though much weaker than a fully annotated corpus, these forms still require non-trivial, corpus- specific knowledge from experts. For example, nominating seed words requires experts to consider (a) NYT-Topics (b) NYT-Locations Figure 1: Visualizations of the same news corpus using Average BERT Representations on two criteria. Colors denote different classes. their relevance to not only the desired classes but also the input corpus; To acquire a few labeled doc- uments per class, unless the classes are balanced, one needs to sample and annotate a much larger number of documents to cover the minority class. In this paper, we focus on extremely weak su- pervision, i.e., only relying on the surface text of class names. This setting is much more challeng- ing than the ones above, and can be considered as almost-unsupervised text classification. We opt to attack this problem from a represen- tation learning perspective—ideal document rep- resentations should lead to nearly the same result between clustering …

Content truncated to 500 words