Where to Go for the Holidays: Towards Mixed-Type Dialogs for Clarification of User Goals

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 1024 - 1034 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Where to Go for the Holidays: Towards Mixed-Type Dialogs for Clarification of User Goals Zeming Liu1∗ †, Jun Xu2∗ ‡, Zeyang Lei2, Haifeng Wang2, Zheng-Yu Niu2, Hua Wu2 1Research Center for Social Computing and Information Retrieval, Harbin Institute of Technology, Harbin, China 2Baidu Inc., Beijing, China zmliu@ir.hit.edu.cn, {xujun03, leizeyang, wanghaifeng, niuzhengyu, wu_hua}@baidu.com Abstract Most dialog systems posit that users have fig- ured out clear and specific goals before start- ing an interaction. For example, users have determined the departure, the destination, and the travel time for booking a flight. However, in many scenarios, limited by experience and knowledge, users may know what they need, but still struggle to figure out clear and specific goals by determining all the necessary slots. In this paper, we identify this challenge, and make a step forward by collecting a new human-to-human mixed-type dialog corpus. It contains 5k dialog sessions and 168k utter- ances for 4 dialog types and 5 domains. Within each session, an agent first provides user-goal- related knowledge to help figure out clear and specific goals, and then help achieve them. Furthermore, we propose a mixed-type dia- log model with a novel Prompt-based con- tinual learning mechanism. Specifically, the mechanism enables the model to continually strengthen its ability on any specific type by utilizing existing dialog corpora effectively. 1 Introduction One of the overarching goals of Artificial Intel- ligence is to build an intelligent agent that can generate coherent multi-turn dialogs to meet user needs/goals. Recently, multiple dialog agents have been launched, such as Echo and Siri. These agents usually position themselves as some kind of “do engines” that act under users’ clear instructions. Specifically, they posit users have figured out clear and specific goals by determining all the necessary aspects or slots of their goals. For example, be- fore booking a flight, a user has determined the departure, the destination and the travel time. ∗Equal contribution. † Mainly responsible for dataset collection during his internship at Baidu. ‡Corresponding author: Jun Xu. However, such assumption can not hold in many real-world scenarios. For example, a user wants to plan a trip to Beijing for relaxing, but he or she only has limited knowledge about Beijing. Thus it is difficult for him or her to decide which slots are needed to achieve this goal. Obviously, in this scene, the user needs additional consultant services from an agent to help figure out clear and specific goals. However, the aforementioned assumption hinders providing these services effectively. In this paper, we make a step towards solving the challenge. In order to facilitate the study of how to help users clarifying their goals, we construct a new Dialog corpus at Baidu, denoted as DuClar- ifyDial. 1 As shown in Figure 1, a user chats about “feels anxious” because of work pressure, and wants to relax himself or herself but …

Content truncated to 500 words