Datasets and Experimental Results
COVID-19-related Tweets Dataset Statistics
Negative
Neutral
Positive
Total
Train
15398
7712
18046
41156
Test
1633
619
1546
3798
Economic texts Dataset Statistics
Negative
Neutral
2 Positive
Total
Train
483
2302
1091
3876
Test
121
576
272
969
E-commerce texts Dataset Statistics
Household
Books
C&A
Electronics
Total
Train
15449
9456
6936
8497
40338
Test
3863
2364
1734
2124
10085
SMS Spam collection Statistics
Normal
Spam
Total
Train
3859
598
4457
Test
966
149
1115
S: with few shot strategy; F: with fine-tuned strategy
COVID-19-RELATED TWEETS Sentiment classification results
Model
ACC($\uparrow$ )
F1($\uparrow$ )
U/E($\downarrow$ )
MNB
0.4037
0.3827
-
LR
0.3875
0.3131
-
RF
0.4462
0.3633
-
DT
0.4037
0.3416
-
KNN
0.3825
0.3481
-
-------------
-----------------
----------------
-------------------
GRU
0.6913
0.6324
-
LSTM
0.6687
0.6312
-
RNN
0.6600
0.6332
-
-------------
-----------------
----------------
-------------------
BART
0.5138
0.3638
-
DeBERTa
0.5375
0.3804
-
-------------
-----------------
----------------
-------------------
GPT-3.5
0.5550
0.5435
0.0000
GPT-4
0.5100
0.5054
0.0000
Gemini-pro
0.5025
0.5105
0.0388
Llama-3-8B
0.5112
0.5149
0.0013
Qwen-7B
0.4913
0.4689
0.0025
Qwen-14B
0.4562
0.4569
0.0100
Vicuna-7B
0.3600
0.3403
0.0000
Vicuna-13B
0.5050
0.4951
0.0013
-------------
-----------------
----------------
-------------------
Gemini-pro(S)
0.4888(-0.014)
0.4880(-0.022)
0.0375(-0.001)
Llama-3-8B(S)
0.5363(+0.025)
0.5298(+0.015)
0.0000(-0.001)
Qwen-7B(S)
0.3900(-0.101)
0.3519(-0.117)
0.0150(+0.012)
Qwen-14B(S)
0.4575(+0.001)
0.4556(-0.001)
0.0037(-0.006)
Vicuna-7B(S)
0.3700(+0.010)
0.3362(-0.004)
0.0013(+0.001)
Vicuna-13B(S)
0.5050(+0.000)
0.4951(+0.000)
0.0000(-0.001)
-------------
-----------------
----------------
-------------------
Llama-3-8B(F)
0.4675(-0.044)
0.4910(-0.024)
0.1175(+0.116)
Qwen-7B(F)
0.8388(+0.348)
0.8433(+0.374)
0.0000(+0.000)
E-Commercial Product Text Classification Results
Model
ACC($\uparrow$ )
F1($\uparrow$ )
U/E($\downarrow$ )
MNB
0.2562
0.2384
-
LR
0.3825
0.2873
-
RF
0.4875
0.3958
-
DT
0.4263
0.4165
-
KNN
0.3762
0.3414
-
-------------
-----------------
----------------
-------------------
GRU
0.9387
0.9383
-
LSTM
0.9363
0.9398
-
RNN
0.8975
0.9010
-
-------------
-----------------
----------------
-------------------
BART
0.7175
0.7246
-
DeBERTa
0.6025
0.6121
-
-------------
-----------------
----------------
-------------------
GPT-3.5
0.9125
0.9152
0.0063
GPT-4
0.9137
0.9221
0.0088
Gemini-pro
0.8775
0.8873
0.0100
Llama-3-8B
0.9113
0.9112
0.0000
Qwen-7B
0.5850
0.6584
0.1850
Qwen-14B
0.6575
0.6843
0.0800
Vicuna-7B
0.7100
0.7164
0.0050
Vicuna-13B
0.8363
0.8503
0.0138
-------------
-----------------
----------------
-------------------
Gemini-pro(S)
0.8862(+0.009)
0.8963(+0.009)
0.0100(+0.000)
Llama-3-8B(S)
0.9062(-0.005)
0.9065(-0.005)
0.0000(+0.000)
Qwen-7B(S)
0.6737(+0.089)
0.8226(+0.164)
0.1812(-0.004)
Qwen-14B(S)
0.7887(+0.131)
0.8548(+0.170)
0.0775(-0.003)
Vicuna-7B(S)
0.7925(+0.083)
0.7899(+0.074)
0.0000(-0.005)
Vicuna-13B(S)
0.9075(+0.071)
0.9153(+0.065)
0.0088(-0.005)
-------------
-----------------
----------------
-------------------
Llama-3-8B(F)
0.9175(+0.006)
0.9164(+0.003)
0.0000(+0.000)
Qwen-7B(F)
0.9713(+0.386)
0.9713(+0.313)
0.0000(-0.185)
ECONOMIC TEXTS Sentiment Classification Results
Model
ACC($\uparrow$ )
F1($\uparrow$ )
U/E($\downarrow$ )
MNB
0.2600
0.2570
-
LR
0.5962
0.3055
-
RF
0.6375
0.4048
-
DT
0.4813
0.3805
-
KNN
0.5325
0.3528
-
-------------
-----------------
----------------
-------------------
GRU
0.6837
0.5494
-
LSTM
0.6950
0.5967
-
RNN
0.6550
0.4298
-
-------------
-----------------
----------------
-------------------
BART
0.4125
0.4152
-
DeBERTa
0.4025
0.4119
-
-------------
-----------------
----------------
-------------------
GPT-3.5
0.6175
0.6063
0.0000
GPT-4
0.7638
0.7659
0.0000
Gemini-pro
0.7488
0.7519
0.0013
Llama-3-8B
0.7675
0.7710
0.0013
Qwen-7B
0.7550
0.7585
0.0025
Qwen-14B
0.7850
0.7860
0.0050
Vicuna-7B
0.7425
0.7250
0.0000
Vicuna-13B
0.6750
0.6735
0.0013
-------------
-----------------
----------------
-------------------
Gemini-pro(S)
0.6925(-0.056)
0.7217(-0.030)
0.0400(+0.039)
Llama-3-8B(S)
0.7550(-0.012)
0.7585(-0.013)
0.0013(+0.000)
Qwen-7B(S)
0.6837(-0.071)
0.6900(-0.069)
0.0288(+0.026)
Qwen-14B(S)
0.7738(-0.011)
0.7748(-0.011)
0.0063(+0.001)
Vicuna-7B(S)
0.7738(+0.031)
0.7607(+0.036)
0.0000(+0.000)
Vicuna-13B(S)
0.7575(+0.082)
0.7616(+0.088)
0.0013(+0.000)
-------------
-----------------
----------------
-------------------
Llama-3-8B
0.7913(+0.024)
0.7796(+0.009)
0.0000(-0.001)
Qwen-7B(F)
0.8400(+0.085)
0.8302(+0.074)
0.0000(-0.003)
SMS SPAM COLLECTION Classification Results
Model
ACC($\uparrow$ )
F1($\uparrow$ )
U/E($\downarrow$ )
MNB
0.7488
0.6376
-
LR
0.8575
0.5419
-
RF
0.8962
0.7196
-
DT
0.8287
0.6559
-
KNN
0.8237
0.6241
-
-------------
-----------------
----------------
-------------------
GRU
0.9675
0.9257
-
LSTM
0.9675
0.9237
-
RNN
0.9725
0.9366
-
-------------
-----------------
----------------
-------------------
BART
0.7137
0.4943
-
DeBERTa
0.7025
0.5630
-
-------------
-----------------
----------------
-------------------
GPT-3.5
0.4988
0.5601
0.0000
GPT-4
0.9463
0.9495
0.0000
Gemini-pro
0.6500
0.7395
0.0575
Llama-3-8B
0.3937
0.4426
0.0025
Qwen-7B
0.7050
0.7527
0.0013
Qwen-14B
0.9137
0.9208
0.0000
Vicuna-7B
0.2762
0.2847
0.0000
Vicuna-13B
0.4550
0.5149
0.0000
-------------
-----------------
----------------
-------------------
Gemini-pro(S)
0.8163(+0.166)
0.8759(+0.136)
0.0488(-0.009)
Llama-3-8B(S)
0.5825(+0.189)
0.6482(+0.206)
0.0088(+0.006)
Qwen-7B(S)
0.7525(+0.047)
0.8124(+0.060)
0.0362(+0.035)
Qwen-14B(S)
0.8525(-0.061)
0.8730(-0.048)
0.0025(+0.003)
Vicuna-7B(S)
0.5675(+0.291)
0.6310(+0.346)
0.0013(+0.001)
Vicuna-13B(S)
0.6412(+0.186)
0.6976(+0.183)
0.0000(+0.000)
-------------
-----------------
----------------
-------------------
Llama-3-8B(F)
0.9825(+0.589)
0.9826(+0.540)
0.0000(-0.003)
Qwen-7B(F)
0.9938(+0.289)
0.9937(+0.241)
0.0000(+0.000)