对话理解中的槽填充

In this paper, the CRF model and several NN models are proposed to solve slot filling problem. We explore the effect of these models based on different corpus, and sentence internal features and history feature are introduced into these models. Results of experiments show, the effect of CRF and NN models in slot filling bases on the size of corpus. When the corpus is small, the CRF model performs better. When the corpus is big enough, the NN model's effect can exceed the CRF model. Experiments also show that the sentence internal feature and dialogue's history feature can improve the effect of slot filling.