基于度量学习的意图识别和槽填充方法

Intention detection and slot filling based on metric learning

人机对话中小样本学习场景下的意图识别和槽填充,是自然语言处理的一个重要课题。本文采用基于度量学习的方法,通过计算query set中的样本与support set中样本的距离,寻找距离最近的类别样本作为分类标签,同时将两个任务联合进行训练,用以提升模型的效果。从实验结果中可以得出,本文提出的Fine-tune方法,对意图识别和槽填充任务都有一定的帮助和提升。胶囊网络在意图识别中也起到了一定的效果,可以帮助去除一部分无关信息,但对槽填充任务的帮助不明显;而任务自适应的投影网络,可以更好地将不同类的向量分开,提升了两个任务的性能。

Intent detection and slot filling in small sample learning scenarios in human-computer dialogue is an important topic in natural language processing.In this paper,a metric-based learning method was adopted.By calculating the distance between the samples in the Query set and the samples in the Support set,the closest category sample was found as the classification label.Meanwhile,the two tasks were trained jointly to improve the effect of the model.From the experimental results,it can be concluded that our fine-tune method has certain help and improvement for intention detection and slot filling task,and capsule network also plays a good effect in intention identification,which can help remove part of irrelevant information,but the help for slot filling task is not obvious.The task-adaptive projection network can better separate the vectors of different classes and improve the performance of the two tasks.