Mem2Seq: Effectively Incorporating Knowledge Bases into End-to-End\n Task-Oriented Dialog Systems

2018/04/22 by Andrea Madotto, Chien-Sheng Wu, Madotto, Andrea +3 · 3 citations
Computer Science · #Topic Modeling #Speech and dialogue systems #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.1804.08217

Abstract

End-to-end task-oriented dialog systems usually suffer from the challenge of\nincorporating knowledge bases. In this paper, we propose a novel yet simple\nend-to-end differentiable model called memory-to-sequence (Mem2Seq) to address\nthis issue. Mem2Seq is the first neural generative model that combines the\nmulti-hop attention over memories with the idea of pointer network. We\nempirically show how Mem2Seq controls each generation step, and how its\nmulti-hop attention mechanism helps in learning correlations between memories.\nIn addition, our model is quite general without complicated task-specific\ndesigns. As a result, we show that Mem2Seq can be trained faster and attain the\nstate-of-the-art performance on three different task-oriented dialog datasets.\n

Cited by

• Affective Multimodal Agents with Proactive Knowledge Grounding for Emotionally Aligned Marketing Dialogue
• Improving LLM's Attachment to External Knowledge In Dialogue Generation Tasks Through Entity Anonymization
• Dual Knowledge-Enhanced Two-Stage Reasoner for Multimodal Dialog Systems

Related

• Gated End-to-End Memory Networks
• Knowledge-based end-to-end memory networks
• Learning to Memorize in Neural Task-Oriented Dialogue Systems
• End-To-End Memory Networks
• Global-to-local Memory Pointer Networks for Task-Oriented Dialogue