Fine-grained Category Discovery under Coarse-grained supervision with Hierarchical Weighted Self-contrastive Learning

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 1314 - 1323 December 7-11, 2022 ©2022 Association for Computational Linguistics Fine-grained Category Discovery under Coarse-grained supervision with Hierarchical Weighted Self-contrastive Learning Wenbin An1, Feng Tian2, Ping Chen3, Siliang Tang4, Qinghua Zheng2, QianYing Wang5 1 School of Automation Science and Engineering, Xi’an Jiaotong University 2 School of Computer Science and Technology, MOEKLNNS Lab, Xi’an Jiaotong University 3 Department of Engineering, University of Massachusetts Boston 4 College of Computer Science and Technology, Zhejiang University 5 Lenovo Research wenbinan@stu.xjtu.edu.cn, {fengtian,qhzheng}@mail.xjtu.edu.cn ping.chen@umb.edu, siliang@zju.edu.cn, wangqya@lenovo.com Abstract Novel category discovery aims at adapting mod- els trained on known categories to novel cat- egories. Previous works only focus on the scenario where known and novel categories are of the same granularity. In this paper, we investigate a new practical scenario called Fine-grained Category Discovery under Coarse- grained supervision (FCDC). FCDC aims at discovering fine-grained categories with only coarse-grained labeled data, which can adapt models to categories of different granularity from known ones and reduce significant la- beling cost. It is also a challenging task since supervised training on coarse-grained cat- egories tends to focus on inter-class distance (distance between coarse-grained classes) but ignore intra-class distance (distance between fine-grained sub-classes) which is essential for separating fine-grained categories. Con- sidering most current methods cannot trans- fer knowledge from coarse-grained level to fine-grained level, we propose a hierarchical weighted self-contrastive network by build- ing a novel weighted self-contrastive mod- ule and combining it with supervised learn- ing in a hierarchical manner. Extensive ex- periments on public datasets show both ef- fectiveness and efficiency of our model over compared methods. Code and data are avail- able at https://github.com/Lackel/ Hierarchical_Weighted_SCL. 1 Introduction Discovering novel categories based on some known categories has attracted much attention in both Nat- ural Language Processing (Zhang et al., 2021; Zhao et al., 2021) and Computer Vision (Zhong et al., 2021; Han et al., 2019). Previous works assume that novel categories are of the same granularity (or of the same class hierarchy level) as known cate- gories. However, in real-world scenarios, novel cat- egories can be more fine-grained sub-categories of known ones (e.g., sports and tennis). A typical ap- … Arts Sports … Arts Sports Root Baseball FootballTennis Music Movies Painting … Root Class Taxonomy FCDC Coarse-grained labels Fine-grained labels Figure 1: An example of proposed FCDC task (fine- grained label names need to be assigned by experts). plication of this scenario is when data analysts want to perform more fine-grained analysis on data with only coarse-grained annotations, where re-labeling fine-grained categories can be time consuming and labour intensive. For example, in the intent de- tection field, discovering more fine-grained user intents can help to provide better services to cus- tomers, but labeling fine-grained intent categories is often much more difficult than labeling coarse- grained ones, since fine-grained annotation often requires higher expertise. To meet this requirement, we investigate a new scenario named Fine-grained Category Discovery under Coarse-grained supervi- sion (FCDC). As shown in …

Content truncated to 500 words