Dialogue Response Selection with Hierarchical Curriculum Learning

We study the learning of a matching model for dialogue response selection. Motivated by the recent finding that models trained with random negative samples are not ideal in real-world scenarios, we propose a hierarchical curriculum learning framework that trains the matching model in an "easy-to-difficult" scheme. Our learning framework consists of two complementary curricula: (1) corpus-level curriculum (CC); and (2) instance-level curriculum (IC). In CC, the model gradually increases its ability in finding the matching clues between the dialogue context and a response candidate. As for IC, it progressively strengthens the model's ability in identifying the mismatching information between the dialogue context and a response candidate. Empirical studies on three benchmark datasets with three state-of-the-art matching models demonstrate that the proposed learning framework significantly improves the model performance across various evaluation metrics.

PDF Abstract ACL 2021 PDF ACL 2021 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Conversational Response Selection Douban SA-BERT+HCL MAP 0.639 # 5
MRR 0.681 # 3
P@1 0.514 # 2
R10@1 0.330 # 1
R10@2 0.531 # 4
R10@5 0.858 # 5
Conversational Response Selection E-commerce SA-BERT+HCL R10@1 0.721 # 7
R10@2 0.896 # 7
R10@5 0.993 # 4
Conversational Response Selection RRS SA-BERT+HCL R10@1 0.454 # 3
MAP 0.671 # 3
MRR 0.683 # 3
P@1 0.503 # 3
R10@2 0.659 # 3
R10@5 0.917 # 3

Methods


No methods listed for this paper. Add relevant methods here