Multi-Label Few-Shot Learning for Aspect Category Detection

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 6330–6340 August 1–6, 2021. ©2021 Association for Computational Linguistics 6330 Multi-Label Few-Shot Learning for Aspect Category Detection Mengting Hu1 Shiwan Zhao2∗ Honglei Guo2 Chao Xue3† Hang Gao1 Tiegang Gao1 Renhong Cheng1 Zhong Su2 1 Nankai University 2 IBM Research - China 3 JD Explore Academy {mthu, knimet}@mail.nankai.edu.cn, {zhaosw, guohl, suzhong}@cn.ibm.com xuechao19@jd.com, {gaotiegang, chengrh}@nankai.edu.cn Abstract Aspect category detection (ACD) in sentiment analysis aims to identify the aspect categories mentioned in a sentence. In this paper, we formulate ACD in the few-shot learning sce- nario. However, existing few-shot learning ap- proaches mainly focus on single-label predic- tions. These methods can not work well for the ACD task since a sentence may contain multiple aspect categories. Therefore, we pro- pose a multi-label few-shot learning method based on the prototypical network. To allevi- ate the noise, we design two effective attention mechanisms. The support-set attention aims to extract better prototypes by removing irrel- evant aspects. The query-set attention com- putes multiple prototype-specific representa- tions for each query instance, which are then used to compute accurate distances with the corresponding prototypes. To achieve multi- label inference, we further learn a dynamic threshold per instance by a policy network. Ex- tensive experimental results on three datasets demonstrate that the proposed method signifi- cantly outperforms strong baselines. 1 Introduction Aspect category detection (ACD) (Pontiki et al., 2014, 2015) is an important task in sentiment anal- ysis. It aims to identify the aspect categories men- tioned in a given sentence from a predefined set of aspect categories. For example, in the sen- tence “the cheesecake is tasty and the staffs are friendly”, two aspect categories, i.e. food and ser- vice, are mentioned. The performance of existing approaches for the ACD task (Zhou et al., 2015; Schouten et al., 2018; Hu et al., 2019) relies heavily on the scale of the labeled dataset. They usually suffer from limited data and fail to generalize well to novel aspect categories with only a few labeled ∗Shiwan Zhao is the corresponding author. †The work was (partially) done in IBM. (A) room_cleanliness Support set (B) staff_owner Query set (A) and (C) (B) (B) (C) hotel 1) Okay, so it is a cute chain hotel. 2) I really don’t see how people are giving this hotel such high ratings. 1) Hotel is just plain dirty. 2) The owners are extremely smart and worldly 3) Not a typical customer service response, especially from the owner ! 1) I think the salon has problems starting with the owner. 2) The owner is very nice. 1) Cleanliness was great, and the food was really good. 2) People have mentioned, bed bugs on yelp !! Figure 1: Example meta-task in a 3-way 2-shot sce- nario. The words in gray background describe the tar- get aspects of interest, while the words marked by the rectangle are irrelevant aspects, which tend to be noise for this meta-task. …

Content truncated to 500 words