A Response Retrieval Approach for Dialogue Using a Multi-Attentive Transformer

This paper presents our work for the ninth edition of the Dialogue System\nTechnology Challenge (DSTC9). Our solution addresses the track number four:\nSimulated Interactive MultiModal Conversations. The task consists in providing\nan algorithm able to simulate a shopping assistant that supports the user with\nhis/her requests. We address the task of response retrieval, that is the task\nof retrieving the most appropriate agent response from a pool of response\ncandidates. Our approach makes use of a neural architecture based on\ntransformer with a multi-attentive structure that conditions the response of\nthe agent on the request made by the user and on the product the user is\nreferring to. Final experiments on the SIMMC Fashion Dataset show that our\napproach achieves the second best scores on all the retrieval metrics defined\nby the organizers. The source code is available at\nhttps://github.com/D2KLab/dstc9-SIMMC.\n