FineD-Eval: Fine-grained Automatic Dialogue-Level Evaluation

By Chen Zhang, Luis Fernando D'Haro, Qiquan Zhang, Thomas Friedrichs, Haizhou Li. 2022

24 citations recorded by the source corpus.

Recent model-based reference-free metrics for open-domain dialogue evaluation exhibit promising correlations with human judgment. However, they either perform turn-level evaluation or look at a single dialogue quality dimension. One would expect a good evaluation metric to assess multiple quality dimensions at the dialogue level. To this end, we are motivated to propose a multi-dimensional dialogue-level metric, which consists of three sub-metrics with each targeting a specific dimension. The sub-metrics are trained with novel self-supervised objectives and exhibit strong correlations with human judgment for their respective dimensions. Moreover, we explore two approaches to combine the sub-metrics: metric ensemble and multitask learning. Both approaches yield a holistic metric that significantly outperforms individual sub-metrics. Compared to the existing state-of-the-art metric, the combined metrics achieve around 16% relative improvement on average across three high-quality dialogue-level evaluation benchmarks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Evaluation & Benchmarks, Dialogue & Conversational AI

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Pengfei Liu
2. Jinlan Fu
3. See-Kiong Ng
4. Zhengbao Jiang
5. Graham Neubig
6. Yixin Liu
7. Weizhe Yuan
8. Xipeng Qiu

In Computer Vision and Pattern Recognition

1. Pengfei Liu
2. See-Kiong Ng
3. Zhengbao Jiang
4. Jinlan Fu

In Evaluation & Benchmarks

1. See-Kiong Ng
2. Jinlan Fu
3. Pengfei Liu
4. Zhengbao Jiang
5. Yixin Liu

Citations in the corpus

• GPTScore: Evaluate as You Desire (2023, methodology): “In this work, we select the evaluation strategy for a specific task based on previous works (Yuan et al., 2021; Zhang et al., 2022a).”