Filtering Noisy Dialogue Corpora by Connectivity and Content Relatedness
This addresses data quality issues for researchers and practitioners training neural dialogue agents, but it is incremental as it builds on existing linguistic insights.
The paper tackles the problem of noisy dialogue datasets by proposing a scoring method based on connectivity and content relatedness to filter out unacceptable utterance pairs, and it shows that training with filtered data improves neural dialogue agents in response generation.
Large-scale dialogue datasets have recently become available for training neural dialogue agents. However, these datasets have been reported to contain a non-negligible number of unacceptable utterance pairs. In this paper, we propose a method for scoring the quality of utterance pairs in terms of their connectivity and relatedness. The proposed scoring method is designed based on findings widely shared in the dialogue and linguistics research communities. We demonstrate that it has a relatively good correlation with the human judgment of dialogue quality. Furthermore, the method is applied to filter out potentially unacceptable utterance pairs from a large-scale noisy dialogue corpus to ensure its quality. We experimentally confirm that training data filtered by the proposed method improves the quality of neural dialogue agents in response generation.