Fig 1.
A sample chat transcript from the annotated dataset.
Table 1.
Statistics of the developed dataset.
Table 2.
Sentiment label distribution across the annotated dataset.
Fig 2.
End-to-end framework for a two-level proposed hierarchical dialogue manager fused with sentiment (ss).
Fig 3.
The architectural diagram of Intent Classifier (IC) module.
Table 3.
Quantitative analysis of intent classification module.
Fig 4.
The architectural diagram of Slot-Filling (SF) module.
Table 4.
Quantitative analysis of sentiment classification module.
Fig 5.
Learning curve of TR based policies during training for different algorithms.
Fig 6.
Learning curve of various policies during training.
Fig 7.
Performance of the VAs during testing with different measures: (a) User Satisfaction, (b) Avg. Turn.
Table 5.
p-values reported by Welch’s t-test on comparing our proposed SR+TR model with other models.
Fig 8.
Performance of the VAs tested with human evaluators: (a) success rate based on binary marking schema, (b) Distribution of user-ratings based on variable marking schema for SR+TR.
Fig 9.
Performance of the VAs during testing: (a) SR+TR, (b) TR.
Table 6.
Quantitative analysis of slot-filling module.