Table 6 Comparison of LEIA-base, LEIA-large, GPT-3.5, and GPT-4 on a random sample of Vent user test split consisting of 1000 examples per emotion category

Affection

74.48 [72.30,76.57]

75.67 [73.56,77.72]

41.38 [38.09,44.69]

37.43 [34.02,40.78]

Anger

72.92 [70.76,75.01]

72.98 [70.87,75.00]

61.79 [59.16,64.26]

66.82 [64.42,69.17]

Fear

69.01 [66.59,71.35]

70.26 [67.89,72.55]

51.55 [48.53,54.55]

60.86 [58.17,63.48]

Happiness

77.69 [75.52,79.77]

77.58 [75.38,79.64]

67.69 [65.60,69.77]

68.70 [66.61,70.77]

Sadness

67.28 [65.00,69.47]

68.00 [65.73,70.22]

59.94 [57.79,62.11]

64.00 [61.82,66.18]

Average

72.28 [71.04,73.50]

72.90 [71.67,74.11]

56.47 [55.07,57.90]

59.56 [58.20,60.92]