SereTOD2022 Track2 Code Analysis - Task-based Dialogue Systems Challenge for Semi-Supervised and Reinforcement Learning

SereTOD2022 Challenge is jointly organized by the team of Mr. Ou Zhijian from Tsinghua University and Mr. Feng Junlan from China Mobile.Workshop on Dialogue Systems">EMNLP2022 Workshop on Task-Based Dialogue Systems for Semi-Supervised and Reinforcement Learning.The data set is oriented to real business scenarios, the oral dialogue is transcribed, and the complexity is extremely high. The data disclosed in the challenge is of great significance. The NLU of the benchmark code is implemented based on BERT, and the NLG is implemented based on GPT2, which has good learning value.Therefore, we analyze the benchmark method and implementation code of Track2 in this challenge in detail.About task settings and data labeling rules See another blog post.

GITHUB Code Base: https://github.com/SereTOD/SereTOD2022

Table of Contents

0.Data Analysis:

0.1 Statistics