- We are very proud to have the following paper accepted at the SIGIR eCom’22 workshop:
- Phillip Lippe, Pengjie Ren, Hinda Haned, Bart Voorn, and Maarten de Rijke. Simultaneously Improving Utility and User Experience in Task-oriented Dialogue Systems.
- Task-oriented dialogue systems (TDSs) help users achieve a specific task through conversations, e.g., in grocery shopping or at help desks. Dialogue response generation (DRG) is a core TDS component that translates system actions into natural language responses. Methods for DRG in TDSs tend to be template-based or corpus-based. The former fill slots in templates with system actions to produce responses at run-time. The latter generate responses token by token by taking system actions into account. In an e-commerce setting, both approaches have strengths and weaknesses: (i) template-based DRG provides high precision and highly predictable responses but may fail to generate diverse and natural responses, thus hurting the user experience; and (ii) corpus-based DRG is able to generate natural responses but its precision or predictability cannot be guaranteed, thus hurting the utility.
- To improve the user experience of conversational interactions without hurting utility we introduce P2-Net, a prototype-based, paraphrasing neural network. P2-Net enhances the precision and diversity of responses. Instead of generating a response from scratch, P2-Net generates system responses by paraphrasing template-based responses. To guarantee precision, P2-Net learns to separate a response into its semantics, context influence, and paraphrasing noise, and to keep the semantics unchanged during paraphrasing. To boost diversity, P2-Net samples previous conversational utterances as prototypes, from which it can then extract speaking style information.
- We conduct experiments on the MultiWOZ dataset with au- tomatic and human evaluations. P2-Net achieves a significant improvement in diversity while preserving the semantics of responses.
