Emotionally Intelligent Chatbots: A Systematic Literature Review

Table 6

Automatic evaluation metrics.

Evaluation type	Metric	# of studies	Studies
Evaluation of generated responses	BLEU	24	[17, 39, 47, 49–53, 55, 57–59, 64, 65, 68, 69, 72–76, 79, 83]
	Perplexity	20	[17, 36, 48, 49, 51, 52, 54, 55, 60–67, 72, 73, 75, 79]
	Distinct-1 grams; Distinct-2 grams	12	[36, 50, 53, 56–59, 64, 67, 74, 76, 80]
	ROUGE	5	[39, 52, 54, 59, 66]
	METEO	4	[39, 48, 52, 66]
Evaluation of emotions	F1	5	[47, 60, 71, 78, 80]
	Precision	9	[17, 47, 48, 60, 67, 69, 71, 78, 84]
	Recall	8	[47, 48, 60, 67, 69, 71, 78, 84]
	Accuracy	22	[8, 17, 48–52, 55, 57, 58, 61, 66, 68, 71–73, 75, 76, 78, 80, 81, 84]