Spaces:
Running
Running
File size: 9,139 Bytes
a1d17f8 | 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 | """
Topic Validation Module
Validates cluster coherence, keyword overlap, and label consistency
"""
import logging
from typing import List, Dict, Optional
import numpy as np
import pandas as pd
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.feature_extraction.text import TfidfVectorizer
logger = logging.getLogger(__name__)
class TopicValidator:
"""Validates topic quality and consistency."""
@staticmethod
def keyword_overlap_score(
keywords_per_topic: Dict[int, List[str]],
threshold: float = 0.3,
) -> Dict[str, float]:
"""
Calculate keyword overlap between topics.
High overlap may indicate mergeable/redundant topics.
Args:
keywords_per_topic: Dict mapping topic_id to keyword list
threshold: Overlap threshold for warnings
Returns:
Dict with overlap scores and warnings
"""
if len(keywords_per_topic) < 2:
return {"mean_overlap": 0.0, "warnings": []}
topics = list(keywords_per_topic.keys())
topic_lists = list(keywords_per_topic.values())
overlaps = []
warnings = []
for i, (tid1, keywords1) in enumerate(zip(topics, topic_lists)):
set1 = set(keywords1)
for tid2, keywords2 in zip(topics[i + 1:], topic_lists[i + 1:]):
set2 = set(keywords2)
if not set1 or not set2:
overlap = 0.0
else:
overlap = len(set1.intersection(set2)) / max(len(set1), len(set2))
overlaps.append(overlap)
if overlap > threshold:
warnings.append(
f"Topics {tid1} and {tid2} share {overlap:.1%} keywords (potential merge)"
)
mean_overlap = np.mean(overlaps) if overlaps else 0.0
return {
"mean_overlap": float(mean_overlap),
"high_overlap_pairs": len(warnings),
"warnings": warnings,
}
@staticmethod
def cluster_coherence(
embeddings: np.ndarray,
labels: np.ndarray,
) -> Dict[str, float]:
"""
Calculate silhouette-like coherence for clusters.
Higher = more separated, more coherent clusters.
Args:
embeddings: Document embeddings (n_docs, n_dims)
labels: Cluster labels (n_docs,)
Returns:
Dict with coherence scores
"""
if len(embeddings) < 2:
return {"mean_coherence": 0.0, "min_coherence": 0.0, "max_coherence": 0.0}
unique_labels = np.unique(labels)
coherence_scores = []
for label in unique_labels:
mask = labels == label
cluster_embeddings = embeddings[mask]
if len(cluster_embeddings) < 2:
coherence_scores.append(0.0)
continue
# Mean pairwise similarity within cluster
similarities = cosine_similarity(cluster_embeddings)
# Average of upper triangle (excluding diagonal)
n = len(similarities)
upper_triangle = similarities[np.triu_indices(n, k=1)]
mean_sim = np.mean(upper_triangle) if len(upper_triangle) > 0 else 0.0
coherence_scores.append(float(mean_sim))
return {
"mean_coherence": float(np.mean(coherence_scores)),
"min_coherence": float(np.min(coherence_scores)),
"max_coherence": float(np.max(coherence_scores)),
"std_coherence": float(np.std(coherence_scores)),
}
@staticmethod
def label_consistency(
topic_labels: Dict[int, str],
topic_keywords: Dict[int, List[str]],
) -> Dict[str, any]:
"""
Check label consistency with keywords.
Labels should be semantically related to their keywords.
Args:
topic_labels: Dict mapping topic_id to label
topic_keywords: Dict mapping topic_id to keyword list
Returns:
Dict with consistency metrics
"""
if not topic_labels:
return {"consistent": True, "issues": []}
issues = []
# Check for empty or very short labels
for tid, label in topic_labels.items():
label_text = str(label).strip()
if not label_text or len(label_text) < 3:
issues.append(f"Topic {tid}: Label too short or empty ('{label_text}')")
# Check label length (should be 3-6 words typically)
word_count = len(label_text.split())
if word_count < 2:
issues.append(f"Topic {tid}: Label too short ({word_count} word)")
elif word_count > 10:
issues.append(f"Topic {tid}: Label too long ({word_count} words)")
# Check for common issues
if "topic" in label_text.lower() and "unknown" in label_text.lower():
issues.append(f"Topic {tid}: Generic/fallback label '{label}'")
consistency_rate = 1.0 - (len(issues) / max(len(topic_labels), 1)) if issues else 1.0
return {
"consistent": len(issues) == 0,
"consistency_rate": float(consistency_rate),
"issue_count": len(issues),
"issues": issues[:10], # Return top 10 issues
}
@staticmethod
def cluster_count_validation(
label_count: int,
min_clusters: int = 15,
max_clusters: int = 30,
) -> Dict[str, any]:
"""
Validate cluster count is within acceptable range.
Args:
label_count: Number of clusters
min_clusters: Minimum acceptable clusters
max_clusters: Maximum acceptable clusters
Returns:
Dict with validation results
"""
valid = min_clusters <= label_count <= max_clusters
status = "β VALID" if valid else "β INVALID"
return {
"valid": valid,
"status": status,
"cluster_count": label_count,
"min_required": min_clusters,
"max_allowed": max_clusters,
"message": (
f"{status}: {label_count} clusters "
f"(target: {min_clusters}-{max_clusters})"
),
}
@staticmethod
def full_validation(
embeddings: np.ndarray,
labels: np.ndarray,
topic_labels: Dict[int, str],
topic_keywords: Dict[int, List[str]],
min_clusters: int = 15,
max_clusters: int = 30,
) -> Dict[str, any]:
"""
Run full validation suite.
Args:
embeddings: Document embeddings
labels: Cluster labels
topic_labels: Topic ID to label mapping
topic_keywords: Topic ID to keywords mapping
min_clusters: Minimum clusters
max_clusters: Maximum clusters
Returns:
Comprehensive validation report
"""
validator = TopicValidator()
return {
"coherence": validator.cluster_coherence(embeddings, labels),
"keyword_overlap": validator.keyword_overlap_score(topic_keywords),
"label_consistency": validator.label_consistency(topic_labels, topic_keywords),
"cluster_count": validator.cluster_count_validation(
len(topic_labels), min_clusters, max_clusters
),
}
@staticmethod
def print_validation_report(validation_result: Dict[str, any]):
"""Pretty print validation report."""
print("\n" + "=" * 70)
print("TOPIC VALIDATION REPORT")
print("=" * 70)
# Coherence
print("\nπ CLUSTER COHERENCE")
coh = validation_result.get("coherence", {})
print(f" Mean Coherence: {coh.get('mean_coherence', 0):.3f} (0.0-1.0)")
print(f" Range: {coh.get('min_coherence', 0):.3f} - {coh.get('max_coherence', 0):.3f}")
# Keyword Overlap
print("\nπ KEYWORD OVERLAP")
overlap = validation_result.get("keyword_overlap", {})
print(f" Mean Overlap: {overlap.get('mean_overlap', 0):.1%}")
high_pairs = overlap.get("high_overlap_pairs", 0)
if high_pairs > 0:
print(f" β οΈ {high_pairs} high-overlap pairs (potential merges)")
for warning in overlap.get("warnings", [])[:3]:
print(f" β {warning}")
# Label Consistency
print("\nβ LABEL CONSISTENCY")
consistency = validation_result.get("label_consistency", {})
rate = consistency.get("consistency_rate", 0)
print(f" Consistency Rate: {rate:.1%}")
issues = consistency.get("issues", [])
if issues:
print(f" Issues Found: {len(issues)}")
for issue in issues[:3]:
print(f" β {issue}")
# Cluster Count
print("\nπ CLUSTER COUNT VALIDATION")
cc = validation_result.get("cluster_count", {})
print(f" {cc.get('message', 'N/A')}")
print("\n" + "=" * 70 + "\n")
|