An Empirical Study on the Overlapping Problem of Open-Domain Dialogue Datasets
This work highlights a data integrity issue for researchers using open-domain dialogue benchmarks, though it is incremental as it focuses on dataset cleaning rather than novel methods.
The study identified overlapping dialogues in DailyDialog and OpenSubtitles datasets, which can artificially inflate performance metrics, and addressed this by cleaning the datasets and establishing a data processing procedure.
Open-domain dialogue systems aim to converse with humans through text, and dialogue research has heavily relied on benchmark datasets. In this work, we observe the overlapping problem in DailyDialog and OpenSubtitles, two popular open-domain dialogue benchmark datasets. Our systematic analysis then shows that such overlapping can be exploited to obtain fake state-of-the-art performance. Finally, we address this issue by cleaning these datasets and setting up a proper data processing procedure for future research.