Exploiting domain-slot related keywords description for Few-Shot Cross-Domain Dialogue State Tracking

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2460 - 2465 December 7-11, 2022 ©2022 Association for Computational Linguistics Exploiting domain-slot related keywords description for Few-Shot Cross-Domain Dialogue State Tracking Qixiang Gao1∗, Guanting Dong1∗, Yutao Mou1∗, Liwen Wang1 Chen Zeng1, Daichi Guo1, Mingyang Sun1, Weiran Xu1∗ 1Beijing University of Posts and Telecommunications, Beijing, China {gqx,dongguanting,myt,w_liwen}@bupt.edu.cn {chenzeng,guodaichi,mysun,xuweiran}@bupt.edu.cn Abstract Collecting dialogue data with domain-slot- value labels for dialogue state tracking (DST) could be a costly process. In this paper, we pro- pose a novel framework based on domain-slot related description to tackle the challenge of few-shot cross-domain DST. Specifically, we design an extraction module to extract domain- slot related verbs and nouns in the dialogue. Then, we integrates them into the description, which aims to prompt the model to identify the slot information. Furthermore, we introduce a random sampling strategy to improve the do- main generalization ability of the model. We utilize a pre-trained model to encode contexts and description and generates answers with an auto-regressive manner. Experimental results show that our approaches substantially outper- form the existing few-shot DST methods on MultiWOZ and gain strong improvements on the slot accuracy comparing to existing slot description methods. 1 Introduction Dialogue state tracking (DST) is an essential com- ponent in a task-oriented dialogue system. It aims to keep track of users’ domains, intents and slots information at each turn of the conversation, which helps to provide sufficient information for select- ing the next system operation (Balaraman et al., 2021). Recent neural-based DST models (Wu et al., 2019; Heck et al., 2020; Feng et al., 2020; Rastogi et al., 2020; Ye et al., 2021) have made significant progress under the availability of large-scale la- beled data. However, due to the high cost of data annotation and the lack of sufficient data in some specific domains, the performance of the general model will drop significantly. Therefore, lack of generalization to new domains hinder the further application of task-oriented dialogue systems in practical industrial scenarios. ∗The first three authors contribute equally. Weiran Xu is the corresponding author. User: I want to book at the allenbell for 4 nights for 8 people. User: I want to book at the allenbell for 4 nights for 8 people. T5 Question: What is the book stay of the hotel that the user is interested in? Slottype: Number of stay for the hotel booking. none 8 Human: Length of stay at the hotel. none Stay of the hotel booking which may include {book, like} or {nights},its output type is {number}. T5 4 Slot_v: {book, like} Slot_n: {nights} Previous method Ours dataset Figure 1: The effect of incorporating domain-slot re- lated keywords into the description for Few-shot Cross- Domain DST task. Current works mainly adopt two ways to deal with DST tasks in few-shot cross-domain scenario: (1) Modular methods (Wu et al., 2019; Zhang et al., 2019; Lee et al., 2019; Wang et al., 2020; Kumar et al., 2020; Ouyang et al., 2020; Guo et al., 2021). …

Content truncated to 500 words