yidingp/icm_OpinionQA

Persona Elicitation

Qwen 30 Evaluation Results

Label accuracy by party / persona (merged folds, separate groups)

Interactive Pipeline Results

Repo structure

Results for interactive pipeline (including log files): ./interactive_search_results
Results for evaluation data (with prediction confidence): ./evaluation_data/confidence

Interactive label accuracy

Evaluation results

Accuracy summary tables

Zero-shot (Base)

Fold	Accuracy
1	0.5652
2	0.5733
3	0.5638
4	0.5575

Zero-shot (Instruct)

Fold	Accuracy
1	0.6894
2	0.7467
3	0.7047
4	0.6379

Few-shot Gold (Instruct)

Fold	k=5	k=10	k=15	k=20	k=30	k=50	k=70	k=80
1	0.8696	0.8012	0.8571	0.8447	0.8509	0.8509	0.8261	0.8323
2	0.8933	0.8667	0.8800	0.8533	0.8400	0.8800	0.8533	0.8400
3	0.8389	0.7919	0.7919	0.7987	0.8389	0.8255	0.8389	0.8389
4	0.7586	0.7644	0.7414	0.7701	0.7529	0.7816	0.7586	0.7299

Few-shot ICM (Instruct)

Fold	k=5	k=10	k=15	k=20	k=30	k=50	k=70	k=80
1	0.8137	0.8261	0.8137	0.8385	0.8261	0.8199	0.8199	0.8447
2	0.8533	0.8400	0.8667	0.8800	0.8667	0.8400	0.8800	0.8533
3	0.8255	0.8121	0.8188	0.8322	0.8255	0.8389	0.8121	0.8322
4	0.7356	0.7759	0.7759	0.7931	0.7644	0.7759	0.7701	0.7471

Zero-shot (Base)

Running: python3 utils_calc_acc.py ./zero_shot_base/result_opinionsqa_zero_shot_fold1.json
Accuracy: 0.5652  (91/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    75     4
 True |    66    16
Binary counts: TP=16 TN=75 FP=4 FN=66
Running: python3 utils_calc_acc.py ./zero_shot_base/result_opinionsqa_zero_shot_fold2.json
Accuracy: 0.5733  (43/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    33     4
 True |    28    10
Binary counts: TP=10 TN=33 FP=4 FN=28
Running: python3 utils_calc_acc.py ./zero_shot_base/result_opinionsqa_zero_shot_fold3.json
Accuracy: 0.5638  (84/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    65     9
 True |    56    19
Binary counts: TP=19 TN=65 FP=9 FN=56
Running: python3 utils_calc_acc.py ./zero_shot_base/result_opinionsqa_zero_shot_fold4.json
Accuracy: 0.5575  (97/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    73    14
 True |    63    24
Binary counts: TP=24 TN=73 FP=14 FN=63

Zero-shot (Instruct)

Running: python3 utils_calc_acc.py ./zero_shot_chat/result_opinionsqa_zero_shot_chat_fold1.json
Accuracy: 0.6894  (111/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True   You ?
False |    74     4     1
 True |    45    37     0
  You |     0     0     0
Running: python3 utils_calc_acc.py ./zero_shot_chat/result_opinionsqa_zero_shot_chat_fold2.json
Accuracy: 0.7467  (56/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    36     1
 True |    18    20
Binary counts: TP=20 TN=36 FP=1 FN=18
Running: python3 utils_calc_acc.py ./zero_shot_chat/result_opinionsqa_zero_shot_chat_fold3.json
Accuracy: 0.7047  (105/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    67     7
 True |    37    38
Binary counts: TP=38 TN=67 FP=7 FN=37
Running: python3 utils_calc_acc.py ./zero_shot_chat/result_opinionsqa_zero_shot_chat_fold4.json
Accuracy: 0.6379  (111/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    75    12
 True |    51    36
Binary counts: TP=36 TN=75 FP=12 FN=51

Few-shot gold (Instruct)

Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few30_fold1.json
Accuracy: 0.8509  (137/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    65    14
 True |    10    72
Binary counts: TP=72 TN=65 FP=14 FN=10
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few50_fold1.json
Accuracy: 0.8509  (137/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    63    16
 True |     8    74
Binary counts: TP=74 TN=63 FP=16 FN=8
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few70_fold1.json
Accuracy: 0.8261  (133/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    63    16
 True |    12    70
Binary counts: TP=70 TN=63 FP=16 FN=12
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few80_fold1.json
Accuracy: 0.8323  (134/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    60    19
 True |     8    74
Binary counts: TP=74 TN=60 FP=19 FN=8
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few30_fold2.json
Accuracy: 0.8400  (63/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    28     9
 True |     3    35
Binary counts: TP=35 TN=28 FP=9 FN=3
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few50_fold2.json
Accuracy: 0.8800  (66/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    30     7
 True |     2    36
Binary counts: TP=36 TN=30 FP=7 FN=2
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few70_fold2.json
Accuracy: 0.8533  (64/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    29     8
 True |     3    35
Binary counts: TP=35 TN=29 FP=8 FN=3
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few80_fold2.json
Accuracy: 0.8400  (63/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    30     7
 True |     5    33
Binary counts: TP=33 TN=30 FP=7 FN=5
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few30_fold3.json
Accuracy: 0.8389  (125/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    61    13
 True |    11    64
Binary counts: TP=64 TN=61 FP=13 FN=11
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few50_fold3.json
Accuracy: 0.8255  (123/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    60    14
 True |    12    63
Binary counts: TP=63 TN=60 FP=14 FN=12
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few70_fold3.json
Accuracy: 0.8389  (125/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    60    14
 True |    10    65
Binary counts: TP=65 TN=60 FP=14 FN=10
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few80_fold3.json
Accuracy: 0.8389  (125/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    60    14
 True |    10    65
Binary counts: TP=65 TN=60 FP=14 FN=10
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few30_fold4.json
Accuracy: 0.7529  (131/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    62    25
 True |    18    69
Binary counts: TP=69 TN=62 FP=25 FN=18
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few50_fold4.json
Accuracy: 0.7816  (136/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    60    27
 True |    11    76
Binary counts: TP=76 TN=60 FP=27 FN=11
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few70_fold4.json
Accuracy: 0.7586  (132/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    59    28
 True |    14    73
Binary counts: TP=73 TN=59 FP=28 FN=14
Running: python3 utils_calc_acc.py ./few_shot_gold_instruct/results_gold_few80_fold4.json
Accuracy: 0.7299  (127/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    54    33
 True |    14    73
Binary counts: TP=73 TN=54 FP=33 FN=14

Few-shot icm (Instruct)

Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few30_fold1.json
Accuracy: 0.8261  (133/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    63    16
 True |    12    70
Binary counts: TP=70 TN=63 FP=16 FN=12
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few50_fold1.json
Accuracy: 0.8199  (132/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    59    20
 True |     9    73
Binary counts: TP=73 TN=59 FP=20 FN=9
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few70_fold1.json
Accuracy: 0.8199  (132/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    61    18
 True |    11    71
Binary counts: TP=71 TN=61 FP=18 FN=11
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few80_fold1.json
Accuracy: 0.8447  (136/161)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    62    17
 True |     8    74
Binary counts: TP=74 TN=62 FP=17 FN=8
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few30_fold2.json
Accuracy: 0.8667  (65/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    31     6
 True |     4    34
Binary counts: TP=34 TN=31 FP=6 FN=4
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few50_fold2.json
Accuracy: 0.8400  (63/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    29     8
 True |     4    34
Binary counts: TP=34 TN=29 FP=8 FN=4
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few70_fold2.json
Accuracy: 0.8800  (66/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    31     6
 True |     3    35
Binary counts: TP=35 TN=31 FP=6 FN=3
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few80_fold2.json
Accuracy: 0.8533  (64/75)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    29     8
 True |     3    35
Binary counts: TP=35 TN=29 FP=8 FN=3
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few30_fold3.json
Accuracy: 0.8255  (123/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    61    13
 True |    13    62
Binary counts: TP=62 TN=61 FP=13 FN=13
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few50_fold3.json
Accuracy: 0.8389  (125/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    62    12
 True |    12    63
Binary counts: TP=63 TN=62 FP=12 FN=12
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few70_fold3.json
Accuracy: 0.8121  (121/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    59    15
 True |    13    62
Binary counts: TP=62 TN=59 FP=15 FN=13
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few80_fold3.json
Accuracy: 0.8322  (124/149)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    60    14
 True |    11    64
Binary counts: TP=64 TN=60 FP=14 FN=11
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few30_fold4.json
Accuracy: 0.7644  (133/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    59    28
 True |    13    74
Binary counts: TP=74 TN=59 FP=28 FN=13
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few50_fold4.json
Accuracy: 0.7759  (135/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    61    26
 True |    13    74
Binary counts: TP=74 TN=61 FP=26 FN=13
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few70_fold4.json
Accuracy: 0.7701  (134/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    59    28
 True |    12    75
Binary counts: TP=75 TN=59 FP=28 FN=12
Running: python3 utils_calc_acc.py ./few_shot_icm_instruct/results_icm_few80_fold4.json
Accuracy: 0.7471  (130/174)  | skipped: 0
Confusion matrix (rows=gold, cols=pred):
        False  True
False |    57    30
 True |    14    73
Binary counts: TP=73 TN=57 FP=30 FN=14

Evaluation data

--- Processing Fold 1 ---
Generated: fold1_test_opinionsqa.json (Samples: 161)
Generated: fold1_train_icm_opinionsqa.json (Samples: 398)
Generated: fold1_train_gold_opinionsqa.json (Samples: 398)

--- Processing Fold 2 ---
Generated: fold2_test_opinionsqa.json (Samples: 75)
Generated: fold2_train_icm_opinionsqa.json (Samples: 484)
Generated: fold2_train_gold_opinionsqa.json (Samples: 484)

--- Processing Fold 3 ---
Generated: fold3_test_opinionsqa.json (Samples: 149)
Generated: fold3_train_icm_opinionsqa.json (Samples: 410)
Generated: fold3_train_gold_opinionsqa.json (Samples: 410)

--- Processing Fold 4 ---
Generated: fold4_test_opinionsqa.json (Samples: 174)
Generated: fold4_train_icm_opinionsqa.json (Samples: 385)
Generated: fold4_train_gold_opinionsqa.json (Samples: 385)

icm searching results summary

Democrat

File	Samples	Accuracy
1	66	0.89
2	53	0.852
3	55	0.891
4	61	0.919

Independent

File	Samples	Accuracy
1	38	0.763
2	8	0.444
3	53	0.852
4	54	0.727

Republican

File	Samples	Accuracy
1	58	1.0
2	14	1.0
3	41	0.81
4	59	0.85