Task-oriented dialogue (TOD) systems are designed to assist users to accomplish their goals, and have gained more and more attention recently in both academia and
industry with the current advances in neural approaches. A TOD system typically consists of several modules, which track user goals to update dialog states, query a
task-related knowledge base (KB) using the dialog states, decide actions and generate responses. Unfortunately, building TOD systems remains a label-intensive, timeconsuming task for two main reasons. First, training neural TOD systems requires manually labeled dialog states and system acts (if used), in both traditional modular approach and recent end-to-end trainable approach. Second, it is often assumed that a task-related knowledge base is available. But for system development from scratch in many real-world tasks, expert labors are needed to construct the KB from annotating unstructured data. Thus, the labeled-data scarcity challenge hinders efficient development of TOD systems at scale.
Therefore,we introduce a new shared task "Towards Semi-Supervised and Reinforced Task-Oriented Dialog Systems (SereTOD)", aiming to benchmark semi-supervised task-oriented dialog systems, built for automated customer-service for mobile operators. The task consists of two tracks: information extraction from dialog transcripts (Track 1) and semi-supervised task oriented dialog systems (Track 2).
Organizers: Zhijian Ou, Junlan Feng, Juanzi Li, Yakun Li, Hong Liu, Hao Peng, Yi Huang, Jiangjiang Zhao
• Apr 29, 2022 Registration opening for shared task
• Jun 30, 2022 Registration deadline for shared task
• Jul 01, 2022 Training data release
• Aug 14, 2022 Evaluation data release
• Aug 21, 2022 Entry submission deadline
• Aug 31, 2022 Evaluation results announced
• Sept 07, 2022 Workshop paper due
• Oct 09, 2022 Notification of paper acceptance
• Track Proposal
• Track 1 Details
• Track 2 Details
• The participating teams can choose to participate in Track 1 independently, and the scores will be ranked according to the performance of information extraction.
• When teams choose to participate in Track 2, they can use the Track 1 system provided by the organizing committee or developed by themselves. The ranking is based on the performance of both objective and manual evaluation.
• The challenge website: http://seretod.org/Challenge.html. Teams can Download and fill in the application form and send it to the designated email address: [email protected].
• The organizing committee will review the qualification of the participating teams. Successful teams will sign the Challenge Data Usage Agreement and be eligible to participate in the challenge.
• Download channels for data will be provided to approved individuals or teams.
For specific enquiries about Challenge (Track1 or Track2)
Please feel free to contact: [email protected]