Table 10 Performance comparison of intermediate models on out-of-domain datasets

BERTweet-base

52.40 [47.01,57.99]

47.16 [44.74,49.61]

43.68 [41.90,45.41]

69.70 [67.14,72.25]

67.23 [63.30,71.06]

2.60

BERTweet-base+eMLM

52.07 [46.85,57.61]

46.26 [43.89,48.66]

42.43 [40.64,44.16]

70.45 [67.90,72.97]

74.79 [71.24,78.22]

2.00

LEIA-base

54.18 [48.79,59.88]

46.31 [43.98,48.72]

43.87 [42.05,45.61]

71.68 [69.18,74.19]

70.37 [66.63,74.01]

1.40

BERTweet-large

51.87 [46.77,57.32]

45.51 [43.14,47.94]

44.08 [42.28,45.85]

68.75 [66.20,71.31]

80.49 [77.18,83.72]

2.40

BERTweet-large+eMLM

52.28 [47.01,57.79]

46.94 [44.58,49.31]

43.78 [41.98,45.57]

69.02 [66.47,71.55]

80.09 [76.77,83.37]

2.00

LEIA-large

54.17 [48.68,59.84]

45.75 [43.45,48.09]

44.12 [42.34,45.89]

70.04 [67.48,72.52]

79.94 [76.69,83.14]

1.60