File size: 10,639 Bytes
ba1d61a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 | import json
import re
import ast
from sklearn.metrics import accuracy_score, f1_score, classification_report, confusion_matrix
from collections import Counter, defaultdict
from datetime import datetime
import numpy as np
def extract_stance_from_output(model_output):
if not model_output:
return None, False, "empty_output"
valid_stances = ['support', 'refute', 'comment', 'unrelated']
try:
if "{'stance':" in model_output or '{"stance":' in model_output:
cleaned_output = model_output.strip()
json_match = re.search(r'\{[^}]*\}', cleaned_output)
if json_match:
cleaned_output = json_match.group()
try:
parsed = ast.literal_eval(cleaned_output)
except:
parsed = json.loads(cleaned_output)
if 'stance' in parsed and isinstance(parsed['stance'], str):
stance = parsed['stance'].lower().strip()
if stance in valid_stances:
return stance, True, None
else:
return stance, False, "invalid_stance_label"
cleaned_output = model_output.lower().strip()
for stance in valid_stances:
if cleaned_output == stance:
return stance, True, None
for stance in valid_stances:
if stance in cleaned_output:
return stance, False, "stance_found_in_text"
if 'oppose' in cleaned_output or 'against' in cleaned_output or 'deny' in cleaned_output:
return 'refute', False, "inferred_from_synonyms"
elif 'agree' in cleaned_output or 'favor' in cleaned_output or 'endorse' in cleaned_output:
return 'support', False, "inferred_from_synonyms"
elif 'neutral' in cleaned_output or 'discuss' in cleaned_output or 'mention' in cleaned_output:
return 'comment', False, "inferred_from_synonyms"
elif 'irrelevant' in cleaned_output or 'unconnected' in cleaned_output or 'off-topic' in cleaned_output:
return 'unrelated', False, "inferred_from_synonyms"
return None, False, "no_stance_pattern"
except Exception as e:
return None, False, f"parsing_error_{str(e)}"
def parse_ground_truth_stance(ground_truth):
if isinstance(ground_truth, dict):
return ground_truth.get('stance', '').lower().strip()
elif isinstance(ground_truth, str):
try:
parsed = ast.literal_eval(ground_truth)
if isinstance(parsed, dict) and 'stance' in parsed:
return parsed['stance'].lower().strip()
except:
pass
return ground_truth.lower().strip()
else:
return ''
def evaluate_multimodal_stance_detection(result_file_path):
with open(result_file_path, 'r', encoding='utf-8') as f:
results = json.load(f)
predictions = []
ground_truths = []
detailed_results = []
extraction_errors = defaultdict(list)
prediction_errors = defaultdict(list)
stance_labels = ['support', 'refute', 'comment', 'unrelated']
for item in results:
item_id = item['id']
model_output = item['model_output']
gt_stance = parse_ground_truth_stance(item['ground_truth'])
pred_stance, is_valid, error_type = extract_stance_from_output(model_output)
detailed_item = {
'id': item_id,
'model_output': model_output,
'extracted_prediction': pred_stance,
'ground_truth': gt_stance,
'correct': pred_stance == gt_stance if pred_stance else False,
'valid': is_valid
}
detailed_results.append(detailed_item)
if not is_valid:
extraction_errors[error_type].append(item_id)
elif pred_stance != gt_stance:
error_pattern = f"{gt_stance}_to_{pred_stance}"
prediction_errors[error_pattern].append(item_id)
if pred_stance:
predictions.append(pred_stance)
ground_truths.append(gt_stance)
if len(predictions) == 0:
return {
'error': 'No valid predictions found',
'total_samples': len(results),
'extraction_errors': dict(extraction_errors)
}
accuracy = accuracy_score(ground_truths, predictions)
micro_f1 = f1_score(ground_truths, predictions, average='micro')
macro_f1 = f1_score(ground_truths, predictions, average='macro')
weighted_f1 = f1_score(ground_truths, predictions, average='weighted')
cm = confusion_matrix(ground_truths, predictions, labels=stance_labels)
class_report = classification_report(ground_truths, predictions,
target_names=stance_labels,
output_dict=True,
zero_division=0)
per_class_metrics = {}
for i, label in enumerate(stance_labels):
if label in class_report:
true_positives = cm[i, i]
false_positives = np.sum(cm[:, i]) - true_positives
false_negatives = np.sum(cm[i, :]) - true_positives
per_class_metrics[label] = {
'precision': round(class_report[label]['precision'], 4),
'recall': round(class_report[label]['recall'], 4),
'f1_score': round(class_report[label]['f1-score'], 4),
'support': int(class_report[label]['support']),
'true_positives': int(true_positives),
'false_positives': int(false_positives),
'false_negatives': int(false_negatives)
}
support_refute_predictions = []
support_refute_ground_truths = []
for pred, true in zip(predictions, ground_truths):
if true in ['support', 'refute']:
support_refute_predictions.append(pred if pred in ['support', 'refute'] else 'other')
support_refute_ground_truths.append(true)
support_refute_accuracy = 0
if len(support_refute_ground_truths) > 0:
support_refute_accuracy = accuracy_score(support_refute_ground_truths, support_refute_predictions)
evaluation_result = {
'task_info': {
'task_name': 'multimodal.stance.detection',
'dataset': 'MMWTWT',
'evaluation_time': datetime.now().isoformat(),
'total_samples': len(results),
'valid_predictions': len(predictions),
'extraction_success_rate': round(len(predictions) / len(results), 4)
},
'metrics': {
'ACC': round(accuracy, 4),
'Micro_F1': round(micro_f1, 4),
'Macro_F1': round(macro_f1, 4),
'Weighted_F1': round(weighted_f1, 4),
'Support_Refute_ACC': round(support_refute_accuracy, 4)
},
'per_class_metrics': per_class_metrics,
'confusion_matrix': {
'labels': stance_labels,
'matrix': cm.tolist(),
'normalized': (cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]).round(4).tolist()
},
'error_analysis': {
'extraction_errors': {
error_type: {
'count': len(sample_ids),
'sample_ids': sample_ids
} for error_type, sample_ids in extraction_errors.items()
},
'prediction_errors': {
error_pattern: {
'count': len(sample_ids),
'sample_ids': sample_ids
} for error_pattern, sample_ids in prediction_errors.items()
}
},
'distribution': {
'ground_truth': dict(Counter(ground_truths)),
'predictions': dict(Counter(predictions))
},
'stance_analysis': {
'support_vs_refute_samples': len(support_refute_ground_truths),
'comment_samples': ground_truths.count('comment'),
'unrelated_samples': ground_truths.count('unrelated'),
'most_confused_pairs': []
}
}
confusion_pairs = []
for i, label1 in enumerate(stance_labels):
for j, label2 in enumerate(stance_labels):
if i != j and cm[i, j] > 0:
confusion_pairs.append({
'true_stance': label1,
'predicted_stance': label2,
'count': int(cm[i, j]),
'percentage': round(cm[i, j] / np.sum(cm[i, :]) * 100, 2)
})
confusion_pairs.sort(key=lambda x: x['count'], reverse=True)
evaluation_result['stance_analysis']['most_confused_pairs'] = confusion_pairs[:5]
base_name = result_file_path.replace('.json', '')
eval_output_file = f"{base_name}_evaluation.json"
with open(eval_output_file, 'w', encoding='utf-8') as f:
json.dump(evaluation_result, f, ensure_ascii=False, indent=2)
detailed_output_file = f"{base_name}_detailed_results.json"
with open(detailed_output_file, 'w', encoding='utf-8') as f:
json.dump(detailed_results, f, ensure_ascii=False, indent=2)
problem_samples = [item for item in detailed_results if not item['correct']]
if problem_samples:
problem_report_file = f"{base_name}_problem_samples.json"
with open(problem_report_file, 'w', encoding='utf-8') as f:
json.dump(problem_samples, f, ensure_ascii=False, indent=2)
print(f"Evaluation complete: {len(results)} samples")
print(f"Key metrics: ACC={evaluation_result['metrics']['ACC']}, Micro F1={evaluation_result['metrics']['Micro_F1']}")
print(f"Support/Refute accuracy: {evaluation_result['metrics']['Support_Refute_ACC']}")
print(f"Extraction success rate: {evaluation_result['task_info']['extraction_success_rate']}")
print(f"Results saved to: {eval_output_file}")
if problem_samples:
print(f"Problematic samples: {len(problem_samples)}; see {problem_report_file} for details")
if confusion_pairs:
print("\nMost confusable stance pairs:")
for pair in confusion_pairs[:3]:
print(f" {pair['true_stance']} → {pair['predicted_stance']}: {pair['count']} times ({pair['percentage']}%)")
return evaluation_result
if __name__ == "__main__":
result_file = "model_result.json"
try:
evaluation_result = evaluate_multimodal_stance_detection(result_file)
except FileNotFoundError:
print(f"Error: file not found {result_file}")
except json.JSONDecodeError:
print(f"Error: invalid format for {result_file}")
except Exception as e:
print(f"Evaluation failed: {str(e)}")
|