File size: 11,373 Bytes
e962f8e | 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 | """mention prediction을 final article-local graph contract로 조립하는 deterministic 책임.
EventMention/EvidenceSpan은 final node가 아니며 evidence는 source offset property로 남긴다.
Neo4j 저장은 이 모듈의 책임이 아니다.
"""
from __future__ import annotations
from dataclasses import dataclass, field
from datetime import datetime, timedelta
import re
from typing import Iterable, Mapping, Sequence
from ..contracts import LEGACY_RELATION_ONTOLOGY
from ..policies.constraints import (
validate_cardinality,
validate_hard_relation_endpoint,
validate_subevent_semantics,
)
from .coreference import UnionFindClusterer
@dataclass(frozen=True, slots=True)
class EvidenceProperty:
article_id: str
start: int
end: int
text: str
sentence_index: int
def validate(self, content: str) -> None:
if not 0 <= self.start < self.end <= len(content):
raise ValueError("evidence offset is outside source text")
if content[self.start : self.end] != self.text:
raise ValueError("evidence text does not round-trip to immutable source")
@dataclass(frozen=True, slots=True)
class GraphNode:
node_id: str
kind: str
properties: Mapping[str, object]
evidence: tuple[EvidenceProperty, ...] = ()
@dataclass(frozen=True, slots=True)
class GraphEdge:
edge_type: str
source_id: str
target_id: str
confidence: float
evidence: tuple[EvidenceProperty, ...] = ()
properties: Mapping[str, object] = field(default_factory=dict)
@dataclass(frozen=True, slots=True)
class ArticleGraph:
article_id: str
nodes: tuple[GraphNode, ...]
edges: tuple[GraphEdge, ...]
class RelativeTimeNormalizer:
"""상대 표현만 publishedAt을 anchor로 정규화한다; 무조건 Event 날짜 fallback은 없다."""
_DAY_OFFSETS = {"그제": -2, "어제": -1, "오늘": 0, "내일": 1, "모레": 2}
_YEAR_OFFSETS = {"작년": -1, "지난해": -1, "올해": 0, "내년": 1}
def normalize(self, text: str, published_at: str | None) -> str | None:
if published_at is None:
return None
anchor = _parse_datetime(published_at)
for word, offset in self._DAY_OFFSETS.items():
if word in text:
return (anchor + timedelta(days=offset)).date().isoformat()
for word, offset in self._YEAR_OFFSETS.items():
if word in text:
return f"{anchor.year + offset:04d}"
match = re.fullmatch(r"(\d{4})년", text.strip())
return match.group(1) if match else None
class DeterministicGraphAssembler:
"""clustering, edge 승격, OCCURRED_ON, dedupe/self-edge/schema validation을 수행한다."""
def __init__(self, relation_ontology_mode: str = LEGACY_RELATION_ONTOLOGY) -> None:
self.clusterer = UnionFindClusterer()
self.time_normalizer = RelativeTimeNormalizer()
self.relation_ontology_mode = relation_ontology_mode
def assemble(
self,
*,
article_id: str,
content: str,
published_at: str | None,
entity_mentions: Sequence[GraphNode],
event_mentions: Sequence[GraphNode],
statements: Sequence[GraphNode],
time_expressions: Sequence[GraphNode],
story_nodes: Sequence[GraphNode] = (),
entity_merge_edges: Iterable[tuple[str, str, float]] = (),
event_merge_edges: Iterable[tuple[str, str, float]] = (),
mention_edges: Iterable[GraphEdge] = (),
merge_threshold: float = 0.5,
) -> ArticleGraph:
all_evidence = [
evidence
for node in (
*entity_mentions,
*event_mentions,
*statements,
*time_expressions,
*story_nodes,
)
for evidence in node.evidence
]
for evidence in all_evidence:
if evidence.article_id != article_id:
raise ValueError("cross-article evidence is forbidden")
evidence.validate(content)
entity_clusters = self.clusterer.cluster(
(node.node_id for node in entity_mentions), entity_merge_edges, threshold=merge_threshold
)
event_clusters = self.clusterer.cluster(
(node.node_id for node in event_mentions), event_merge_edges, threshold=merge_threshold
)
entity_map, entity_nodes = _materialize_clusters("ENTITY", entity_clusters, entity_mentions)
event_map, event_nodes = _materialize_clusters("EVENT", event_clusters, event_mentions)
time_nodes = tuple(
GraphNode(
node.node_id,
"TIME",
{
**dict(node.properties),
"normalized": node.properties.get("normalized")
or self.time_normalizer.normalize(
str(node.properties.get("text", node.evidence[0].text if node.evidence else "")),
published_at,
),
},
node.evidence,
)
for node in time_expressions
)
final_nodes = tuple(
(*entity_nodes, *event_nodes, *statements, *time_nodes, *story_nodes)
)
node_by_id = {node.node_id: node for node in final_nodes}
node_ids = set(node_by_id)
lifted = []
for edge in mention_edges:
source = entity_map.get(edge.source_id, event_map.get(edge.source_id, edge.source_id))
target = entity_map.get(edge.target_id, event_map.get(edge.target_id, edge.target_id))
if edge.edge_type in {"CAUSES", "RESPONDS_TO", "SUBEVENT_OF"} and source == target:
continue
if source not in node_ids or target not in node_ids:
raise ValueError("edge references an unknown final node")
validate_cardinality(edge.edge_type, source, target)
validate_hard_relation_endpoint(
edge.edge_type,
node_by_id[source].kind,
node_by_id[target].kind,
ontology_mode=self.relation_ontology_mode,
)
if edge.edge_type == "SUBEVENT_OF":
validate_subevent_semantics(
node_by_id[target].properties,
edge.properties,
)
lifted.append(
GraphEdge(
edge.edge_type,
source,
target,
edge.confidence,
edge.evidence,
edge.properties,
)
)
if edge.edge_type == "TIME":
lifted.append(
GraphEdge(
"OCCURRED_ON",
source,
target,
edge.confidence,
edge.evidence,
edge.properties,
)
)
edges = _deduplicate_edges(lifted)
graph = ArticleGraph(
article_id,
final_nodes,
edges,
)
self.validate(graph)
return graph
def validate(self, graph: ArticleGraph) -> None:
ids = [node.node_id for node in graph.nodes]
if len(ids) != len(set(ids)):
raise ValueError("duplicate final graph node ID")
known = set(ids)
node_by_id = {node.node_id: node for node in graph.nodes}
for edge in graph.edges:
if edge.source_id not in known or edge.target_id not in known:
raise ValueError("graph edge endpoint is absent")
validate_cardinality(edge.edge_type, edge.source_id, edge.target_id)
validate_hard_relation_endpoint(
edge.edge_type,
node_by_id[edge.source_id].kind,
node_by_id[edge.target_id].kind,
ontology_mode=self.relation_ontology_mode,
)
if edge.edge_type == "SUBEVENT_OF":
validate_subevent_semantics(
node_by_id[edge.target_id].properties,
edge.properties,
)
_validate_subevent_hierarchy(graph.edges, maximum_depth=2)
def _materialize_clusters(
kind: str,
clusters: Sequence[Sequence[str]],
mentions: Sequence[GraphNode],
) -> tuple[dict[str, str], tuple[GraphNode, ...]]:
by_id = {node.node_id: node for node in mentions}
mapping: dict[str, str] = {}
nodes = []
for index, members in enumerate(clusters, start=1):
cluster_id = f"{kind.lower()}-{index:04d}"
evidence = tuple(item for member in members for item in by_id[member].evidence)
properties: dict[str, object] = {"mention_ids": tuple(members)}
if kind == "EVENT":
member_nodes = [by_id[member] for member in members]
properties.update(
{
"grounded": all(
node.properties.get("grounded") is True
for node in member_nodes
),
"bounded_episode": all(
node.properties.get("bounded_episode") is True
for node in member_nodes
),
"synthetic": any(
node.properties.get("synthetic") is True
for node in member_nodes
),
}
)
for member in members:
mapping[member] = cluster_id
nodes.append(
GraphNode(
cluster_id,
kind,
properties,
evidence,
)
)
return mapping, tuple(nodes)
def _deduplicate_edges(edges: Iterable[GraphEdge]) -> tuple[GraphEdge, ...]:
best: dict[tuple[str, str, str], GraphEdge] = {}
for edge in edges:
key = edge.edge_type, edge.source_id, edge.target_id
if key not in best or edge.confidence > best[key].confidence:
best[key] = edge
return tuple(best[key] for key in sorted(best))
def _validate_subevent_hierarchy(
edges: Iterable[GraphEdge], *, maximum_depth: int
) -> None:
"""child→parent forest에서 cycle과 초기 depth 상한을 검사한다."""
parents: dict[str, set[str]] = {}
for edge in edges:
if edge.edge_type == "SUBEVENT_OF":
parents.setdefault(edge.source_id, set()).add(edge.target_id)
def depth(node: str, path: tuple[str, ...]) -> int:
if node in path:
raise ValueError("SUBEVENT_OF cycle is forbidden")
next_nodes = parents.get(node, ())
if not next_nodes:
return 0
return 1 + max(depth(parent, (*path, node)) for parent in next_nodes)
for node in parents:
if depth(node, ()) > maximum_depth:
raise ValueError(
f"SUBEVENT_OF hierarchy depth exceeds initial limit {maximum_depth}"
)
def _parse_datetime(value: str) -> datetime:
canonical = value.replace("Z", "+00:00")
try:
return datetime.fromisoformat(canonical)
except ValueError as error:
raise ValueError("publishedAt must be ISO-8601 for relative time normalization") from error
|