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
|
import numpy as np
from rouge_score import rouge_scorer
from bert_score import score as bert_score
from typing import Dict, List, Tuple
import difflib
class MeetingMinutesEvaluator:
def __init__(self):
self.rouge_scorer = rouge_scorer.RougeScorer(
['rouge1', 'rouge2', 'rougeL'],
use_stemmer=True
)
def evaluate_summary(
self,
predicted: str,
reference: str
) -> Dict[str, float]:
"""评估摘要质量"""
rouge_scores = self.rouge_scorer.score(reference, predicted)
P, R, F1 = bert_score(
[predicted],
[reference],
lang="zh",
verbose=False
)
return {
"rouge1_f1": rouge_scores['rouge1'].fmeasure,
"rouge2_f1": rouge_scores['rouge2'].fmeasure,
"rougeL_f1": rouge_scores['rougeL'].fmeasure,
"bert_score_f1": F1.item(),
"combined_score": (
rouge_scores['rougeL'].fmeasure * 0.4 +
F1.item() * 0.6
)
}
def evaluate_key_points(
self,
predicted: List[str],
reference: List[str]
) -> Dict[str, float]:
"""评估关键要点提取"""
matched_count = 0
total_similarity = 0.0
for ref_point in reference:
best_match_score = 0.0
for pred_point in predicted:
similarity = difflib.SequenceMatcher(
None,
ref_point,
pred_point
).ratio()
best_match_score = max(best_match_score, similarity)
if best_match_score > 0.6:
matched_count += 1
total_similarity += best_match_score
recall = matched_count / len(reference) if reference else 0
precision = matched_count / len(predicted) if predicted else 0
f1 = (
2 * precision * recall / (precision + recall)
if (precision + recall) > 0 else 0
)
return {
"precision": precision,
"recall": recall,
"f1_score": f1,
"avg_similarity": total_similarity / len(reference) if reference else 0
}
def evaluate_action_items(
self,
predicted: List[ActionItem],
reference: List[ActionItem]
) -> Dict[str, float]:
"""评估行动项识别"""
pred_descriptions = [item.description for item in predicted]
ref_descriptions = [item.description for item in reference]
base_score = self.evaluate_key_points(
pred_descriptions,
ref_descriptions
)
structure_scores = []
for pred_item in predicted:
best_match = None
best_similarity = 0
for ref_item in reference:
sim = difflib.SequenceMatcher(
None,
pred_item.description,
ref_item.description
).ratio()
if sim > best_similarity:
best_similarity = sim
best_match = ref_item
if best_match and best_similarity > 0.6:
assignee_match = pred_item.assignee == best_match.assignee
priority_match = pred_item.priority == best_match.priority
structure_scores.append(
(assignee_match + priority_match) / 2
)
structure_accuracy = (
np.mean(structure_scores) if structure_scores else 0
)
return {
**base_score,
"structure_accuracy": structure_accuracy,
"weighted_f1": base_score['f1_score'] * 0.7 + structure_accuracy * 0.3
}
def evaluate_full_output(
self,
predicted: GroundTruth,
reference: GroundTruth
) -> Dict[str, any]:
"""完整评估"""
results = {
"meeting_id": reference.meeting_id,
"summary_scores": self.evaluate_summary(
predicted.summary,
reference.summary
),
"key_points_scores": self.evaluate_key_points(
predicted.key_points,
reference.key_points
),
"action_items_scores": self.evaluate_action_items(
predicted.action_items,
reference.action_items
)
}
results["overall_score"] = (
results["summary_scores"]["combined_score"] * 0.30 +
results["key_points_scores"]["f1_score"] * 0.35 +
results["action_items_scores"]["weighted_f1"] * 0.35
)
return results
|
