Dynamic Semantic Matching and Aggregation Network for Few-shot Intent Detection

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1209–1218 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1209 Dynamic Semantic Matching and Aggregation Network for Few-shot Intent Detection Hoang Nguyen1, Chenwei Zhang2, Congying Xia1, Philip S. Yu1 1 Department of Computer Science, University of Illinois at Chicago, Chicago, IL, USA 2 Amazon, Seattle, WA, USA {hnguy7,cxia8,psyu}@uic.edu, cwzhang@amazon.com Abstract Few-shot Intent Detection is challenging due to the scarcity of available annotated utter- ances. Although recent works demonstrate that multi-level matching plays an impor- tant role in transferring learned knowledge from seen training classes to novel testing classes, they rely on a static similarity measure and overly fine-grained matching components. These limitations inhibit generalizing capabil- ity towards Generalized Few-shot Learning settings where both seen and novel classes are co-existent. In this paper, we propose a novel Semantic Matching and Aggregation Network where semantic components are distilled from utterances via multi-head self-attention with additional dynamic regularization constraints. These semantic components capture high-level information, resulting in more effective match- ing between instances. Our multi-perspective matching method provides a comprehensive matching measure to enhance representations of both labeled and unlabeled instances. We also propose a more challenging evaluation set- ting that considers classification on the joint all-class label space. Extensive experimental results demonstrate the effectiveness of our method. Our code and data are publicly avail- able 1 . 1 Introduction Intent Detection (ID) is a crucial task in natural lan- guage understanding, whose objective is to extract underlying intents behind the given utterances. The extracted intents could provide further contexts for further downstream Natural Language Processing tasks such as dialogue state tracking or question answering. Unlike traditional text classification, ID is challenging for two main reasons (1) Utter- ances are usually short and diversely expressed, 1https://github.com/nhhoang96/ Semantic_Matching (2) Emerging intents occur continuously, especially across different domains (Liu et al., 2019a). Despite recent advances, state-of-the-art ID methods (Haihong et al., 2019; Goo et al., 2018) require a large amount of annotated data to achieve competitive performance. This requirement in- hibits models’ capability in generalizing to newly emerging intents with no or limited annotations during inference. Re-training or fine-tuning large models on few samples of emerging classes could easily lead to overfitting problems. Motivated by human capability in correctly cat- egorizing new classes with only a few examples (Lake et al., 2011; Gidaris and Komodakis, 2018), few-shot learning (FSL) paradigms are adopted to tackle the scarcity problems of emerging classes. FSL methods take advantage of a small set of la- beled examples (support set) to learn how to dis- criminate unlabeled samples (query samples) be- tween classes, even those not seen during training. Recent works in FSL (Sun et al., 2019; Ye and Ling, 2019) focus on learning the matching infor- mation between the labeled samples (support) and the unlabeled samples (query) to provide additional contextual information for instance-level represen- tations, leading to effective prototype representa- tion. However, these methods only extract similar- ity based on fine-grained word semantics, failing to capture the …

Content truncated to 500 words