Spaces:
Sleeping
Sleeping
File size: 16,372 Bytes
5644c15 1f7b841 5644c15 1f7b841 5644c15 1f7b841 5644c15 1f7b841 5644c15 1f7b841 5644c15 1f7b841 5644c15 1f7b841 5644c15 1f7b841 5644c15 4e9058c 5644c15 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 | import networkx as nx
import requests
from typing import Dict, List, Optional, Tuple
class KnowledgeGraph:
"""Manages the knowledge graph for image-concept relationships"""
def __init__(self):
self.graph = nx.MultiDiGraph()
self.conceptnet_api_base = "http://api.conceptnet.io"
self.relationship_weights = {
'IsA': 1.0,
'HasProperty': 0.8,
'RelatedTo': 0.7,
'PartOf': 0.9,
'UsedFor': 0.8,
'CapableOf': 0.8,
'AtLocation': 0.7,
'default': 0.5
}
self.decay_factor = 0.8 # Weight decay for depth
def get_relationship_weight(self, relationship: str, confidence: float = 1.0) -> float:
"""
Calculate edge weight based on relationship type and confidence
"""
base_weight = self.relationship_weights.get(relationship,
self.relationship_weights['default'])
return base_weight * confidence
def add_weighted_relationships(self, source: str, relationships: List[Dict]) -> None:
"""
Enhanced version of add_relationships with confidence scores
"""
for rel in relationships:
# Extract confidence from ConceptNet response
confidence = rel.get('weight', 1.0)
rel_type = rel['relationship']
# Calculate weight
weight = self.get_relationship_weight(rel_type, confidence)
# Add nodes and weighted edge
for node in [rel['source'], rel['target']]:
if not self.graph.has_node(node):
self.graph.add_node(node, node_type='concept')
self.graph.add_edge(
rel['source'],
rel['target'],
relationship=rel_type,
weight=weight,
confidence=confidence
)
def add_image_node(self, image_id: str, caption: str) -> None:
"""
Add an image node with its caption to the graph
Args:
image_id: Unique identifier for the image
caption: BLIP-2 generated caption for the image
"""
# Add image node with its properties
self.graph.add_node(
image_id,
node_type='image',
caption=caption
)
# Extract main concepts from caption (simple tokenization for now)
concepts = [word.lower() for word in caption.split()]
# Add edges between image and its concepts
for concept in concepts:
self.graph.add_node(
concept,
node_type='concept'
)
self.graph.add_edge(
image_id,
concept,
relationship='has_concept'
)
# def expand_concept(self, concept: str) -> List[Dict]:
# """
# Query ConceptNet for relationships about a concept
# Args:
# concept: The concept to query relationships for
# Returns:
# List of dictionaries containing relationship data
# """
# # Format concept for ConceptNet API (lowercase, replace spaces with underscores)
# formatted_concept = f"/c/en/{concept.lower().replace(' ', '_')}"
# try:
# # Query ConceptNet API
# response = requests.get(
# f"{self.conceptnet_api_base}{formatted_concept}",
# params={'limit': 50} # Adjust limit as needed
# )
# response.raise_for_status()
# # Extract edges (relationships) from response
# edges = response.json().get('edges', [])
# # Filter and format relationships
# relationships = []
# for edge in edges:
# # Only consider English relationships
# if all(lang.endswith('/en') for lang in [edge['start']['language'], edge['end']['language']]):
# relationships.append({
# 'source': edge['start']['label'],
# 'target': edge['end']['label'],
# 'relationship': edge['rel']['label']
# })
# return relationships
# except requests.exceptions.RequestException as e:
# print(f"Error querying ConceptNet: {e}")
# return []
def expand_concept(self, concept: str) -> List[Dict]:
"""Query ConceptNet for relationships about a concept"""
formatted_concept = f"/c/en/{concept.lower().replace(' ', '_')}"
try:
response = requests.get(
f"{self.conceptnet_api_base}{formatted_concept}",
params={'limit': 50}
)
response.raise_for_status()
edges = response.json().get('edges', [])
relationships = []
for edge in edges:
try:
# More robust checking of edge structure
start = edge.get('start', {})
end = edge.get('end', {})
# Check if we have valid English concepts
if (start.get('language', '') == 'en' and
end.get('language', '') == 'en'):
relationships.append({
'source': start.get('label', ''),
'target': end.get('label', ''),
'relationship': edge.get('rel', {}).get('label', '')
})
except (KeyError, TypeError):
continue # Skip malformed edges
return relationships
except requests.exceptions.RequestException as e:
print(f"Error querying ConceptNet: {e}")
return []
def add_relationships(self, source: str, relationships: List[Dict]) -> None:
"""
Add relationships from ConceptNet to our graph
Args:
source: The source concept
relationships: List of relationship dictionaries from expand_concept
"""
for rel in relationships:
# Add source and target nodes if they don't exist
for node in [rel['source'], rel['target']]:
if not self.graph.has_node(node):
self.graph.add_node(
node,
node_type='concept'
)
# Add edge with relationship type
self.graph.add_edge(
rel['source'],
rel['target'],
relationship=rel['relationship'],
weight=1.0 # Default weight, could be adjusted based on ConceptNet's confidence
)
# Add reverse relationship for bidirectional search
self.graph.add_edge(
rel['target'],
rel['source'],
relationship=f"reverse_{rel['relationship']}",
weight=1.0
)
def search(self, query: str, limit: int = 5) -> List[str]:
"""
Search for images based on semantic query
Args:
query: Search query string
limit: Maximum number of results to return
Returns:
List of image IDs ordered by relevance
"""
# First, expand the query concept to understand its relationships
query_relationships = self.expand_concept(query)
# Add query relationships temporarily to graph
temp_query_node = f"_query_{query}"
self.add_relationships(temp_query_node, query_relationships)
try:
# Find all image nodes
image_nodes = [n for n, attr in self.graph.nodes(data=True)
if attr.get('node_type') == 'image']
# Calculate relevance scores for each image
image_scores = []
for image_id in image_nodes:
# Use shortest path length as a relevance metric
# Shorter paths = more relevant
try:
path_length = nx.shortest_path_length(
self.graph,
source=temp_query_node,
target=image_id
)
score = 1.0 / (1.0 + path_length) # Convert distance to similarity score
image_scores.append((image_id, score))
except nx.NetworkXNoPath:
continue
# Sort by score and return top results
image_scores.sort(key=lambda x: x[1], reverse=True)
return [img_id for img_id, _ in image_scores[:limit]]
finally:
# Clean up temporary query nodes
self.graph.remove_node(temp_query_node)
def get_related_concepts(self, concept: str, relationship_type: Optional[str] = None) -> List[Tuple[str, str]]:
"""
Get concepts related to given concept, optionally filtered by relationship type
Args:
concept: The source concept to find relations for
relationship_type: Optional filter for specific relationship types
Returns:
List of tuples containing (related_concept, relationship_type)
"""
related_concepts = []
# Get all outgoing edges from the concept
if self.graph.has_node(concept):
for _, target, edge_data in self.graph.out_edges(concept, data=True):
rel_type = edge_data.get('relationship', '')
# Filter by relationship_type if specified
if relationship_type is None or rel_type == relationship_type:
# Don't include temporary query nodes or image nodes
if (not target.startswith('_query_') and
self.graph.nodes[target].get('node_type') != 'image'):
related_concepts.append((target, rel_type))
# If no direct relationships found, try expanding from ConceptNet
if not related_concepts:
new_relationships = self.expand_concept(concept)
self.add_relationships(concept, new_relationships)
# Try again with newly added relationships
for _, target, edge_data in self.graph.out_edges(concept, data=True):
rel_type = edge_data.get('relationship', '')
if relationship_type is None or rel_type == relationship_type:
if (not target.startswith('_query_') and
self.graph.nodes[target].get('node_type') != 'image'):
related_concepts.append((target, rel_type))
return related_concepts
def search_with_depth(self, query: str, max_depth: int = 3, limit: int = 5) -> List[Tuple[str, float]]:
"""Enhanced search with depth control and path weights"""
temp_query_node = f"_query_{query}"
try:
# First add the query node and its relationships
query_relationships = self.expand_concept(query)
if not query_relationships:
print(f"No relationships found for query: {query}")
return []
# Ensure query node is added before adding relationships
self.graph.add_node(temp_query_node, node_type='query')
self.add_weighted_relationships(temp_query_node, query_relationships)
# Rest of the search logic...
image_scores = {}
image_nodes = [n for n, attr in self.graph.nodes(data=True)
if attr.get('node_type') == 'image']
for image_id in image_nodes:
try:
paths = nx.all_simple_paths(
self.graph,
source=temp_query_node,
target=image_id,
cutoff=max_depth
)
path_scores = []
for path in paths:
score = self._calculate_path_score(path)
path_scores.append(score)
if path_scores:
image_scores[image_id] = max(path_scores)
except nx.NetworkXNoPath:
continue
# Sort and return results
sorted_results = sorted(
image_scores.items(),
key=lambda x: x[1],
reverse=True
)
return sorted_results[:limit]
finally:
# Clean up: remove temporary query node
if self.graph.has_node(temp_query_node):
self.graph.remove_node(temp_query_node)
def search_with_depth1(self, query: str, max_depth: int = 3, limit: int = 5) -> List[Tuple[str, float]]:
"""
Enhanced search with depth control and path weights
Args:
query: Search query string
max_depth: Maximum path length to consider
limit: Maximum number of results to return
Returns:
List of tuples (image_id, relevance_score)
"""
temp_query_node = f"_query_{query}"
query_relationships = self.expand_concept(query)
self.add_weighted_relationships(temp_query_node, query_relationships)
try:
image_scores = {}
# Find all image nodes
image_nodes = [n for n, attr in self.graph.nodes(data=True)
if attr.get('node_type') == 'image']
for image_id in image_nodes:
# Get all paths up to max_depth
try:
paths = nx.all_simple_paths(
self.graph,
source=temp_query_node,
target=image_id,
cutoff=max_depth
)
# Calculate score for each path
path_scores = []
for path in paths:
score = self._calculate_path_score(path)
path_scores.append(score)
# Use maximum score from all paths
if path_scores:
image_scores[image_id] = max(path_scores)
except nx.NetworkXNoPath:
continue
# Sort and return top results
sorted_results = sorted(
image_scores.items(),
key=lambda x: x[1],
reverse=True
)
return sorted_results[:limit]
finally:
self.graph.remove_node(temp_query_node)
def _calculate_path_score(self, path: List[str]) -> float:
"""
Calculate score for a path based on relationship weights and depth
"""
total_score = 0
path_length = len(path) - 1
for i in range(path_length):
# Get edge data (might have multiple edges between nodes)
edges = self.graph.get_edge_data(path[i], path[i + 1])
if edges:
# Use maximum weight among parallel edges
max_weight = max(
edge.get('weight', self.relationship_weights['default'])
for edge in edges.values()
)
# Apply depth decay
depth_factor = self.decay_factor ** i
total_score += max_weight * depth_factor
# Normalize by path length
return total_score / path_length if path_length > 0 else 0
|