Overview

Task-oriented dialog (TOD) systems are designed to assist users to accomplish their goals. Recently, neural generative approaches have received increasing attention. Unfortunately, building TOD systems remains as a label-intensive and time-consuming task. The process still heavily relies on manually labeled dialog data and annotated task-related knowledge base. However, unlabeled data are often easily available in many forms such as human-to-human dialogs, open-domain text corpus, and unstructured knowledge documents. Moreover, the nature of sequential decision making is not fully considered in current TOD systems. The purpose of this workshop is to invite researchers from both academia and industry to share their perspectives on building semi-supervised and reinforced TOD systems, discuss challenges and advance the field in joint effort. In parallel, a challenge is organized to foster this line of research, with a newly released, large-scale, human-human dialog dataset, called the MobileCS (Mobile Customer Service) dataset, which consists of 100K real-world dialogs. The challenge includes two tracks: Information extraction from dialog transcripts (Track 1), and Task-oriented dialog systems (Track 2).

Check out SereTOD

News

2025/01/07  To facilitate future research, we re-open the challenge, and people who are interested in building systems for the challenge task can register and send email to FutureDialRAG (at) gmail (dot) com to obtain the data. The end date is open and the organizers reserve the right as to when to end the challenge.

2023/01/06  The Proceedings of the SereTOD Workshop can be accessed here. The system description papers from the top three teams for Track 1 and Track 2 of SereTOD Challenge respectively can be found here.

2022/12/17  The live recording of the SereTOD Workshop, including all sessions of Opening Remarks, 3 Invited Talks, 4 Oral Sessions, SereTOD Challenge Awards, and Panel, could be accessed here! You can select different sessions by the video selector on the right of the video. Enjoy it!

2022/11/27  Due to recent situation regarding COVID-19 in Beijing, EMNLP 2022 SereTOD Workshop is to be held fully virtually via a Zoom webinar. The virtual poster session can be accessed from EMNLP 2022 online platform via GatherTown. Click here to join the Workshop!

2022/11/23  Information about invited speakers and panel posted! SereTOD 2022 Panel Questionnaire opened up! Feel free to let us know what topics you would like to see discussed at the panel.

2022/11/03  SereTOD Workshop Program published!

2022/10/10  Notification of paper acceptance sent out! Given the high-quality submissions received and the capacity of the Workshop, the selection process was very competitive and the acceptance rate is 47%, with 11 accepted papers.

2022/09/14  The ground-truth labels/responses of the test data for both Track 1 and Track 2 have been released via email.

2022/09/10  We are pleased to announce the evaluation results for both Track 1 and Track 2, which can be accessed via this link. Congratulations to the 15 teams, who submitted effective results, out of the total of 62 teams registered for SereTOD Challenge!

2022/08/27  Manuscripts that have been reviewed for EMNLP 2022 and were rejected have the opportunity to be considered for publication in the Workshop proceedings. The deadline for submitting EMNLP manuscripts and reviews is October 7, 2022. See details in Call for Papers.

2022/08/14  Entry Submission due date is extended by 10 days to August 31. Accordingly, the due dates for Evaluation Results Announced and Regular Paper Submission are also extended by 10 days.

2022/08/14  Evaluation data have been released to registered teams.

2022/08/02  We have provided computer resources to 12 teams. Thanks the support from JiuTian !

2022/07/05  The code repository for baseline systems and evaluation scripts for both Track 1 and Track 2 are released.

2022/07/01  The MobileCS dataset for SereTOD Challenge has been released to registered teams.

2022/04/29  SereTOD Challenge Kickoff Seminar is completed! Introduction to the Challenge (by Zhijian Ou), talk videos from invited speakers (from Xiaodong He, Luo Si, Kai Yu, Zhengyu Niu, Wanxiang Che), panel recording (hosted by Junlan Feng), are available here !

2022/04/19  Registration opening for challenge, as detailed in Call for Participation !

2022/02/23  Announcement of Call for Papers !

Important Dates (AOE)

July 1, 2022 Registration deadline for challenge (extended)
July 1, 2022 Training data release (extended)
July 15, 2022 ACL Rolling Review paper due (submit via ARR)
August 14, 2022 Evaluation data release
August 31, 2022 Entry submission deadline (extended)
September 10, 2022 Evaluation results announced (extended)
September 17, 2022 Regular paper submission due (via softconf) (extended)
October 2, 2022 ARR commitment due (via OpenReview)
October 7, 2022 Submission of EMNLP manuscripts with reviews (via softconf)
October 10, 2022 Notification of paper acceptance
October 16, 2022 Camera-ready papers due
December 7, 2022 EMNLP 2022 Workshop Date (virtual)

Invited Speakers

  • Dilek Hakkani-Tur, Amazon

  • Jason Williams, Apple

  • Pascale Fung, HKUST

    • Supported by

    Joint Institute of Tsinghua University - China Mobile Communications Group Co. Ltd.

    Contact

    Zhijian Ou (ozj@tsinghua.edu.cn)

    Junlan Feng (fengjunlan@chinamobile.com)

    Juanzi Li (lijuanzi@tsinghua.edu.cn)

    Copyright © SereTod Organizers. All rights reserved.