In practical dialogue systems, it is crucial to avoid undesired responses and poor user experiences by detecting out-Of-scope (OOS) intents from user utterances. Currently, to detect OOS intents, limited-supervised me...
详细信息
In practical dialogue systems, it is crucial to avoid undesired responses and poor user experiences by detecting out-Of-scope (OOS) intents from user utterances. Currently, to detect OOS intents, limited-supervised methods are more potential due to using limited true OOS data to better model the OOS data distribution. However, existing limited-supervised methods often yield low-quality OOS augmentations, i.e., noisy OOS and long -distant OOS, due to the use of the retrieval-based data augmentation mechanism, which will damage the OOS intentdetection performance. To tackle this problem, in this paper, we propose a novel OOS intentdetection method by leveraging intent-invariant data augmentation, called InInOOS, which can generate high -quality pseudo-OOS utterances with invariant OOS intents but various slot values and different expressions to further enhance the OOS intentdetection performance. Specifically, we first generate pseudo-OOS candidates using our pre-trained intent-invariant utterance generation model, and then the most beneficial candidates are selected to train an OOS intentdetection model. Extensive experiments on two public datasets show that our method performs better than state-of-the-art baselines.
Detecting out-of-scope(OOS) intents in dialogue systems is a challenging technique with practical applications. As for OOS intentdetection, it not only ensures the accuracy of classifying known intents but detecting ...
详细信息
ISBN:
(纸本)9781665488679
Detecting out-of-scope(OOS) intents in dialogue systems is a challenging technique with practical applications. As for OOS intentdetection, it not only ensures the accuracy of classifying known intents but detecting OOS intents is also crucial. Current related models are limited in learning decision boundaries or setting the threshold of confidence score, which all neglect that a well-formed intent representation is a key point. Meanwhile, text extractors trained by traditional cross-entropy loss merely focus on reducing the error rate of the class to which the sample is classified. In this paper, we propose an effective feature extraction method based on deep metric learning to construct the triplet network with prior knowledge. With the constructed triplet loss, mining hard samples, which refers to the far-apart intents between the same class and close intent representations among different classes, can further obtain discriminative intent representations. In addition, we also introduce adversarial training to make intent representations more robust. Experiments on three public datasets prove the effectiveness of our proposed method of learning discriminative intent representations.
The growing demand for trustworthy dialogue systems emphasizes the need for consistently accurate responses to user inputs. The first step in developing a trustworthy dialogue system is detecting user inputs with the ...
详细信息
暂无评论