Beyond Words: An Intelligent Human-Machine Dialogue System with Multimodal Generation and Emotional Comprehension

Table 2

Manual annotation results of different methods (lower values indicating better models).

Ground truth vs	Appropriateness (%)	Informativeness (%)	Emotional (%)
Text-based	57.3	69.3	59.8
Emotion-based	54.2	62.7	54.2
Image-Chat	58.2	55.1	52.8
Ours-Emotion	63.2	61.8	52.1
Ours-Visual	60.4	60.5	56.5
Ours	54.3	53.4	46.7

Bold values indicate the best results.