Guider l'attention dans les modeles de sequence a sequence pour la prediction des actes de dialogue

The task of predicting dialog acts (DA) based on conversational dialog is a\nkey component in the development of conversational agents. Accurately\npredicting DAs requires a precise modeling of both the conversation and the\nglobal tag dependencies. We leverage seq2seq approaches widely adopted in\nNeural Machine Translation (NMT) to improve the modelling of tag sequentiality.\nSeq2seq models are known to learn complex global dependencies while currently\nproposed approaches using linear conditional random fields (CRF) only model\nlocal tag dependencies. In this work, we introduce a seq2seq model tailored for\nDA classification using: a hierarchical encoder, a novel guided attention\nmechanism and beam search applied to both training and inference. Compared to\nthe state of the art our model does not require handcrafted features and is\ntrained end-to-end. Furthermore, the proposed approach achieves an unmatched\naccuracy score of 85% on SwDA, and state-of-the-art accuracy score of 91.6% on\nMRDA.\n