Unknown Intent Detection Based on Large-Margin Cosine Loss

Identifying the unknown (novel) user intents that have never appeared in the training set is a challenging task in the dialogue system. This chapter presents a two-stage method for detecting unknown intents. The method use a bidirectional long short-term memory (BiLSTM) network with the margin loss as the feature extractor. With margin loss, the network is encouraged to maximize inter-class variance and minimize intra-class variance, thereby learning descriminative deep feature. Then, the feature vectors are fed into a density-based novelty detection algorithm, local outlier factor (LOF), to detect unknown intents. Experiments on two benchmark datasets show that the proposed method can yield consistent improvements compared with the baseline methods.