| """Entity relationship graph (journalism suite layer 5). |
| |
| Deterministic proper-noun extraction + co-occurrence edges. Entities that |
| share a sentence repeatedly are surfaced as a RELATIONSHIP for the human to |
| investigate. No model call; no hidden inference. This is the graph half of |
| "who is connected to whom" the suit materializes while the brain reasons. |
| |
| Usage: |
| from research.entitygraph import EntityGraph |
| g = EntityGraph() |
| g.add_doc("s1", "The Central Bank met Delta Corp. Delta Corp hired Smith.") |
| g.report() |
| """ |
| import re |
| from collections import defaultdict |
| from itertools import combinations |
|
|
| _PHRASE = re.compile(r"\b[A-Z][a-zA-Z]{1,25}(?:\s+[A-Z][a-zA-Z]{1,25}){0,3}\b") |
| _ORG_SUFFIX = re.compile(r"\b(?:Inc|Corp|Corporation|Ltd|Agency|Department|" |
| r"Committee|Commission|University|Institute|Bureau|" |
| r"Administration|Bank|Fund|Office|Council|Force|" |
| r"Group|Industries|Systems|Media|News|Post|Times)\b") |
| _STOP = {"The", "This", "That", "These", "Those", "A", "An", "One", "Two", |
| "Mr", "Mrs", "Ms", "Dr", "January", "February", "March", "April", |
| "May", "June", "July", "August", "September", "October", "November", |
| "December", "Monday", "Tuesday", "Wednesday", "Thursday", "Friday", |
| "Saturday", "Sunday"} |
|
|
|
|
| class EntityGraph: |
| def __init__(self): |
| self.docs = {} |
| self.sentences = [] |
|
|
| def add_doc(self, source_id, text): |
| self.docs[source_id] = text |
| for s in re.split(r"(?<=[.!?])\s+", text): |
| s = s.strip() |
| if s: |
| self.sentences.append((source_id, s)) |
|
|
| def extract(self, text): |
| """Proper-noun phrases; drop stop-word-led matches (deterministic).""" |
| out = [] |
| for m in _PHRASE.finditer(text): |
| p = m.group(0).strip() |
| if p.split()[0] in _STOP: |
| continue |
| out.append(p) |
| return sorted(set(out)) |
|
|
| def _nodes(self): |
| nodes = defaultdict(int) |
| for _, sent in self.sentences: |
| for e in self.extract(sent): |
| nodes[e] += 1 |
| return nodes |
|
|
| def edges(self, min_cooccur=2): |
| """Entity pairs sharing a sentence; weight = co-occurrence count.""" |
| pair_w = defaultdict(int) |
| for _, sent in self.sentences: |
| ents = sorted(set(self.extract(sent))) |
| for a, b in combinations(ents, 2): |
| pair_w[(a, b)] += 1 |
| return {k: v for k, v in pair_w.items() if v >= min_cooccur} |
|
|
| def central(self, top=10): |
| """Degree centrality: entities with most distinct graph neighbors.""" |
| deg = defaultdict(int) |
| for (a, b) in self.edges(): |
| deg[a] += 1 |
| deg[b] += 1 |
| return sorted(deg.items(), key=lambda kv: -kv[1])[:top] |
|
|
| def report(self, min_cooccur=2): |
| lines = ["# Entity Relationship Graph", ""] |
| lines.append("## Entities (mentions)") |
| nodes = self._nodes() |
| for e, n in sorted(nodes.items(), key=lambda kv: -kv[1])[:30]: |
| lines.append(f"- {e}: x{n}") |
| lines.append("") |
| lines.append("## Relationships (co-occurrence)") |
| edges = self.edges(min_cooccur) |
| for (a, b), w in sorted(edges.items(), key=lambda kv: -kv[1])[:25]: |
| lines.append(f"- {a} <-> {b} (x{w})") |
| if not edges: |
| lines.append(f"- none above min_cooccur={min_cooccur}") |
| lines.append("") |
| lines.append("## Central entities (degree)") |
| for e, d in self.central(): |
| lines.append(f"- {e}: {d} neighbors") |
| return "\n".join(lines) |
|
|
| def to_dot(self, min_cooccur=2): |
| lines = ["digraph entities {"] |
| for e in self._nodes(): |
| lines.append(f' "{e}";') |
| for (a, b), w in self.edges(min_cooccur).items(): |
| lines.append(f' "{a}" -> "{b}" [label="{w}"];') |
| lines.append("}") |
| return "\n".join(lines) |
|
|
