2020
DOI: 10.1609/aaai.v34i05.6280
|View full text |Cite
|
Sign up to set email alerts
|

Likelihood Ratios and Generative Classifiers for Unsupervised Out-of-Domain Detection in Task Oriented Dialog

Abstract: The task of identifying out-of-domain (OOD) input examples directly at test-time has seen renewed interest recently due to increased real world deployment of models. In this work, we focus on OOD detection for natural language sentence inputs to task-based dialog systems. Our findings are three-fold:First, we curate and release ROSTD (Real Out-of-Domain Sentences From Task-oriented Dialog) - a dataset of 4K OOD examples for the publicly available dataset from (Schuster et al. 2019). In contrast to existing set… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
48
0

Year Published

2021
2021
2024
2024

Publication Types

Select...
3
2
1

Relationship

0
6

Authors

Journals

citations
Cited by 35 publications
(48 citation statements)
references
References 8 publications
0
48
0
Order By: Relevance
“…We selected two baselines to evaluate improvements of our proposed OodGAN. Our baselines for the ROSTD dataset is our implementation of Zheng et al (2020) and the work of Gangal et al (2019). The baseline for the OSQ dataset is our implementation of Zheng et al (2020).…”
Section: Results On Proposed Model Oodganmentioning
confidence: 99%
See 1 more Smart Citation
“…We selected two baselines to evaluate improvements of our proposed OodGAN. Our baselines for the ROSTD dataset is our implementation of Zheng et al (2020) and the work of Gangal et al (2019). The baseline for the OSQ dataset is our implementation of Zheng et al (2020).…”
Section: Results On Proposed Model Oodganmentioning
confidence: 99%
“…They propose three baseline approaches for OOD detection that rely on OOD training data. Gangal et al (2019) created a ROSTD dataset and explored likelihood ratio based approaches. Lee and Shalyminov (2019) proposed an OOD detection method that does not require OOD data by utilizing counterfeit OOD turns in the context of a dialog.…”
Section: Related Workmentioning
confidence: 99%
“…We conduct experiments on two public benchmark datasets after low-resource settings, including CLINC [15], ROSTD [9]. CLINC and ROSTD are for evaluating the performance of intent classification systems in the presence of OOD samples.…”
Section: Setup Datasetsmentioning
confidence: 99%
“…3) LSTM-AutoEncoder [5] trains an autoencoder for OOD detection using only ID data. 4) LR-GC is proposed by [9] to use a likelihood ratio method based on generative classifiers to correct confounding background statistics. 5) ProtoNet [10] is trained only on the naive classification loss L naive .…”
Section: Setup Datasetsmentioning
confidence: 99%
See 1 more Smart Citation