Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Domain Adaptation in Intent Classification Systems: A Review

Jesse Atuhurra, Hidetaka Kamigaito, Taro Watanabe, Eric Nichols

TL;DR

This paper provides a structured survey of domain adaptation in intent classification for dialogue systems, detailing the landscape of datasets, languages, and domains used to train and evaluate ICS. It synthesizes three core methodological approaches—fine-tuning pretrained language models, prompting, and few-shot/zero-shot learning—and highlights representative techniques and datasets that illustrate current capabilities and gaps. The authors identify key limitations in domain adaptation, such as dataset biases, monolingual focus, and the need for multimodal inputs, and offer concrete directions like multilingual, multimodal datasets, conversational pretraining, adapters, and contrastive learning to address these challenges. The work aims to guide future dataset creation and method development to enable robust, domain-diverse, and scalable intent classification in real-world dialogue systems.

Abstract

Dialogue agents, which perform specific tasks, are part of the long-term goal of NLP researchers to build intelligent agents that communicate with humans in natural language. Such systems should adapt easily from one domain to another to assist users in completing tasks. Researchers have developed a broad range of techniques, objectives, and datasets for intent classification to achieve such systems. Despite the progress in developing intent classification systems (ICS), a systematic review of the progress from a technical perspective is yet to be conducted. In effect, important implementation details of intent classification remain restricted and unclear, making it hard for natural language processing (NLP) researchers to develop new methods. To fill this gap, we review contemporary works in intent classification. Specifically, we conduct a thorough technical review of the datasets, domains, tasks, and methods needed to train the intent classification part of dialogue systems. Our structured analysis describes why intent classification is difficult and studies the limitations to domain adaptation while presenting opportunities for future work.

Domain Adaptation in Intent Classification Systems: A Review

TL;DR

This paper provides a structured survey of domain adaptation in intent classification for dialogue systems, detailing the landscape of datasets, languages, and domains used to train and evaluate ICS. It synthesizes three core methodological approaches—fine-tuning pretrained language models, prompting, and few-shot/zero-shot learning—and highlights representative techniques and datasets that illustrate current capabilities and gaps. The authors identify key limitations in domain adaptation, such as dataset biases, monolingual focus, and the need for multimodal inputs, and offer concrete directions like multilingual, multimodal datasets, conversational pretraining, adapters, and contrastive learning to address these challenges. The work aims to guide future dataset creation and method development to enable robust, domain-diverse, and scalable intent classification in real-world dialogue systems.

Abstract

Dialogue agents, which perform specific tasks, are part of the long-term goal of NLP researchers to build intelligent agents that communicate with humans in natural language. Such systems should adapt easily from one domain to another to assist users in completing tasks. Researchers have developed a broad range of techniques, objectives, and datasets for intent classification to achieve such systems. Despite the progress in developing intent classification systems (ICS), a systematic review of the progress from a technical perspective is yet to be conducted. In effect, important implementation details of intent classification remain restricted and unclear, making it hard for natural language processing (NLP) researchers to develop new methods. To fill this gap, we review contemporary works in intent classification. Specifically, we conduct a thorough technical review of the datasets, domains, tasks, and methods needed to train the intent classification part of dialogue systems. Our structured analysis describes why intent classification is difficult and studies the limitations to domain adaptation while presenting opportunities for future work.
Paper Structure (63 sections, 3 equations, 1 figure, 5 tables)

This paper contains 63 sections, 3 equations, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Task definition: Intent Classification
  3. Example
  4. Datasets
  5. NLU and Non-NLU Data
  6. ORCAS-I kusa_alexander_de:vries_2022
  7. Search4Code https://doi.org/10.48550/arxiv.2011.11950
  8. Monolingual and Multilingual Data
  9. Non-English Monolingual Data
  10. Almawave-SLU Bellomaria2019AlmawaveSLUAN
  11. CATSLU 10.1145/3340555.3356098
  12. Vietnamese JointIDSF https://doi.org/10.48550/arxiv.2104.02021
  13. Multilingual Data
  14. Facebook https://doi.org/10.48550/arxiv.1810.13327
  15. MultiATIS++ https://doi.org/10.48550/arxiv.2004.14353
  16. ...and 48 more sections

Figures (1)

  • Figure 1: Design choices for a spoken language understanding (SLU) system https://doi.org/10.48550/arxiv.1904.03670. Top: Conventional ASR → NLU system for SLU. Bottom: End-to-end SLU. At the top, the SLU systems consists of two components: one for automatic speech recognition (ASR) and the other for natural language understanding (NLU). The ASR component maps speech data to textual data. Then, the NLU component maps textual data to intent. At the bottom, one trainable model for end-to-end SLU is used to map speech directly to intent.