Progressive Adversarial Learning for Bootstrapping: A Case Study on Entity Set Expansion

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9673–9682 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9673 Progressive Adversarial Learning for Bootstrapping: A Case Study on Entity Set Expansion Lingyong Yan1,3, Xianpei Han1,2, Le Sun1,2,∗ 1Chinese Information Processing Laboratory 2State Key Laboratory of Computer Science Institute of Software, Chinese Academy of Sciences, Beijing, China 3University of Chinese Academy of Sciences, Beijing, China {lingyong2014, xianpei, sunle}@iscas.ac.cn Abstract Bootstrapping has become the mainstream method for entity set expansion. Conventional bootstrapping methods mostly define the ex- pansion boundary using seed-based distance metrics, which heavily depend on the quality of selected seeds and are hard to be adjusted due to the extremely sparse supervision. In this pa- per, we propose BootstrapGAN, a new learning method for bootstrapping which jointly mod- els the bootstrapping process and the boundary learning process in a GAN framework. Specif- ically, the expansion boundaries of different bootstrapping iterations are learned via differ- ent discriminator networks; the bootstrapping network is the generator to generate new pos- itive entities, and the discriminator networks identify the expansion boundaries by trying to distinguish the generated entities from known positive entities. By iteratively performing the above adversarial learning, the generator and the discriminators can reinforce each other and be progressively refined along the whole boot- strapping process. Experiments show that Boot- strapGAN achieves the new state-of-the-art en- tity set expansion performance. 1 Introduction Bootstrapping is a fundamental technique for en- tity set expansion (ESE). It starts from a few seed entities (e.g., {London, Beijing, Paris}) and iteratively extracts new entities in the target cat- egory (e.g., {Berlin, Moscow, Tokyo}) to expand the entity set, where new entities are often evaluated by their context similarities to seeds (e.g., sharing the same context pattern–“* is an important city”) (Riloff and Jones, 1999; Gupta and Manning, 2014; Yan et al., 2020a). During the above process, it is core to decide whether the new entities belong to the target category (within the expansion bound- ary) or not (outside the expansion boundary) (Shi et al., 2014; Gupta and Manning, 2014). *Corresponding author. Bootstrapping Model Input of iter. k Output of iter. k Ajax U.N. European Union London Paris Beijing Ajax U.N. European Union London Paris Beijing Distance-based boundary Ajax U.N. European Union London Paris Beijing Univ. of Paris Brussels Ajax U.N. European Union London Paris Beijing Univ. of Paris Brussels Unexpanded Entities Unexpanded Entities Correct Expansion Correct Expansion Noisy Expansion Noisy Expansion Seeds Seeds Figure 1: The expansion boundary problem of the boot- strapping technique. The areas in different background colors belong to different categories. Using the distance to positive entities will easily result in a bad expansion boundary at each iteration. However, it is challenging to determine the ex- pansion boundaries during the whole bootstrapping process, since only several seeds are used as the supervision at the beginning. Firstly, it is obviously not enough to define a good boundary using only several positive entities. For example, as shown in Figure 1, when only …

Content truncated to 500 words