Energy-based Self-attentive Learning of Abstractive Communities for Spoken Language Understanding
This work addresses a specific spoken language understanding problem for conversational analysis, with incremental improvements over existing methods.
The paper tackles abstractive community detection in spoken language understanding by grouping utterances that can be summarized by a common abstractive sentence, using a neural contextual utterance encoder with three self-attention mechanisms trained with siamese and triplet energy-based meta-architectures, achieving state-of-the-art performance on the AMI corpus.
Abstractive community detection is an important spoken language understanding task, whose goal is to group utterances in a conversation according to whether they can be jointly summarized by a common abstractive sentence. This paper provides a novel approach to this task. We first introduce a neural contextual utterance encoder featuring three types of self-attention mechanisms. We then train it using the siamese and triplet energy-based meta-architectures. Experiments on the AMI corpus show that our system outperforms multiple energy-based and non-energy based baselines from the state-of-the-art. Code and data are publicly available.