Differences between (a) the supervised DST task and (b) the unsupervised DST task. The strikethrough words are the resources not required by the unsupervised DST task.

Loading [MathJax]/jax/output/SVG/jax.js

Download： Full Size Img PowerPoint

Differences between (a) the supervised DST task and (b) the unsupervised DST task. The strikethrough words are the resources not required by the unsupervised DST task.

2023, 38(4): 834-852.

Unsupervised Dialogue State Tracking for End-to-End Task-Oriented Dialogue with a Multi-Span Prediction Network .

Figures of the Article

Differences between (a) the supervised DST task and (b) the unsupervised DST task. The strikethrough words are the resources not required by the unsupervised DST task.
Architecture of MSPN.
Illustrations of the semantic distance based clustering approach. The red dots denote the slot. The blue dots denote the inferred mentions. The other dots denote the irrelevant word. (a) Nearest slot ranking. (b) Nearest neighbors ranking.
Illustration of the sampling strategy. We first sample a start word and then construct multiple continuous text spans.
Illustration of the training procedure. The robot represents our proposed MSPN.
Case study. The dialogue is in the weather retrieval domain. “G” in the parenthesis denotes the generated dialogue states and the generated responses. “T” denotes the target dialogue states and the target responses.

Others

English PDF
2023-4-8-1064-Highlights 196KB
External link to attachment
https://rdcu.be/dqKC3
DOCX format
2023-4-8-1064-Chinese Information 27KB

Related articles

${\tt{VastPipe}}$ : A High-Throughput Inference System via Adaptive Space-Division Multiplexing for Diverse Accelerators

2025, 40(2): 444-463. DOI: 10.1007/s11390-024-3773-5
A Reinforcement Learning Based Approach to Partition Testing

2025, 40(1): 99-118. DOI: 10.1007/s11390-024-2900-7
HmTest: Automated Testing of HarmonyOS Apps via Model-Driven Navigation and Reinforcement Learning

DOI: 10.1007/s11390-025-5142-4
An Efficient Reinforcement Learning Game Framework for UAV-Enabled Wireless Sensor Network Data Collection

2022, 37(6): 1356-1368. DOI: 10.1007/s11390-022-2419-8
An Exercise Collection Auto-Assembling Framework with Knowledge Tracing and Reinforcement Learning

2022, 37(5): 1105-1117. DOI: 10.1007/s11390-022-2412-2
A Unified Shared-Private Network with Denoising for Dialogue State Tracking

2021, 36(6): 1407-1419. DOI: 10.1007/s11390-020-0338-0
Robustness Assessment of Asynchronous Advantage Actor-Critic Based on Dynamic Skewness and Sparseness Computation: A Parallel Computing View

2021, 36(5): 1002-1021. DOI: 10.1007/s11390-021-1217-z
WATuning: A Workload-Aware Tuning System with Attention-Based Deep Reinforcement Learning

2021, 36(4): 741-761. DOI: 10.1007/s11390-021-1350-8
The Learning Convergence of CMAC in Cyclic Learning

1994, 9(4): 320-328.
Inductive Learning

1993, 8(2): 22-36.