지식 그래프에
말을 걸다

import json
import requests

payload = {
    "content": "Pichette is a partner at Inovia
      Capital and chair of commerce platform
      Lightspeed",
    "lang": "en"
    }
res = requests.post(
  "https://{}/v1/?fields={}&token={}".format
  (HOST, "entities,facts", TOKEN), json=payload)

for ent in res.json()["entities"]:
  print(ent)   # write to graph

for fact in res.json()["facts"]:
  print(fact)  # write to graph

{"sentiment": -0.32881057,
 "entities": [{"name": "Patrick Pichette",
   "confidence": 0.81, "salience": 0.64345313, …}],
 "facts": [{
  "humanReadable": "[Patrick Pichette] employee
     or member of [Innovia Capital]",
  "entity": { "name": "Patrick Pichette",
    "diffbotUri": "…/EYwbLa__MNVGtPMryqqCCEA",
    "confidence": 0.9999268,
    "allUris": ["…wikidata.org/entity/Q3369779"], …},
  "property": {"name": "employee or member of",
    "diffbotUri": "…ontology#Person.memberOf"},
  "value": {"name": "Innovia Capital",
    "confidence": 0.9989188, …},
  "confidence": 0.9498123,
  "evidence": [{"passage": "Pichette is a
     partner at Inovia Capital", …}]}, …]}

# 엔터티–속성–값 삼중항. Wikidata 교차 참조,
# salience·confidence·감성·증거 구절까지 실려 온다

// entity_list, fact_list를 매개변수로 쓰기 트랜잭션 호출
UNWIND $facts AS fact
MERGE (source:Entity { id: fact.entity.diffbotUri })
MERGE (target:Entity { id: fact.value.diffbotUri })
WITH source, target, fact
CALL apoc.create.relationship(source,
    fact.property.name, {}, target)
// salience 등도 함께 싣는다

# 토큰 → 그래프 스키마 요소의 대응표
q1_verb_to_rel = {"direct" : "DIRECTED",
                  "produce" : "PRODUCED",
                  "write" : "WROTE",
                  "review" : "REVIEWED"}

max_match = result[-1]
verb = max_match[1].lemma_               # (1) 동사 추출
title = ' '.join([tk.text for tk in max_match[2:]
         if tk.pos_ != 'PUNCT'])          # (2) 제목 어절 추출
query_as_cypher = "MATCH (p:Person)-[:{rel_type}]->
  (m:Movie) WHERE toLower(m.title) CONTAINS
  '{movie_title}' …".format(
    rel_type=q1_verb_to_rel[verb],
    movie_title=title.lower())           # (3) 합성

MATCH (p:Person)-[:WROTE]->(m:Movie)
WHERE toLower(m.title) CONTAINS 'a few good men'
WITH collect(p.name) AS answer_as_list
RETURN CASE WHEN size(answer_as_list) > 0 THEN
   substring(reduce(result='',
     x in answer_as_list | result + ', ' + x),2)
   ELSE "I cannot answer your question about
     'a few good men' " end AS answer

// 그래프를 눈으로 살폈다면 손수 썼을 바로
// 그 질의다 — 뜻이 통역된 것이다

MERGE (x:Entity { name: "Euroyacht Limited" ,
    jurisdiction: "BM" })
MERGE (y:Entity { name: "TUC LIMITED", jurisdiction: "MLT",
    incorporation_date: date("2013-10-07")})
MERGE (z:Entity {name: "GLOBAL TUITION & EDUCATION
    INSURANCE CORPORATION", jurisdiction: "BRB",
    incorporation_date: date("1998-04-03") })
MERGE (x)-[:OFFICER_OF]->(y)
MERGE (x)-[:OFFICER_OF]->(z)

MATCH (x)-[r]->(y)
RETURN x.name + " " +
  toLower(replace(type(r),"_"," ")) + " " +
  y.name AS sentence

// 결과 — 놀랍도록 올바른 영어에 가깝다:
// "Euroyacht Limited officer of GLOBAL TUITION
//    & EDUCATION INSURANCE CORPORATION"
// "Euroyacht Limited officer of TUC LIMITED"
//
// 스키마 특정 요소가 없는 질의라 어떤 그래프에도
// 유효하다 — :play movies에 돌리면
// "Keanu Reeves acted in Johnny Mnemonic".
// 두 줄 코드치고는 꽤 훌륭한 가성비다

MATCH (n:Entity) UNWIND keys(n) AS property
RETURN n.name + "'s " + replace(property,"_"," ")
  + " is " + n[property] AS sentence

// "Euroyacht Limited's jurisdiction is BM"
// "TUC LIMITED's jurisdiction is MLT"
// "GLOBAL TUITION & … 's incorporation date
//    is 1998-04-03"
// 이번에도 제법 반듯한 영어 문장들이다

@prefix talk: <http://www.neo4j.org/2022/07/talkable#> .
@prefix mv: <http://www.neo4j.org/sch/movies#> .

mv:ACTED_IN rdf:type owl:ObjectProperty ;
   rdfs:domain mv:Person ;
   rdfs:range mv:Movie ;
   talk:direct "acted in"@default ,
               "is in the cast of"@long ,
               "worked in"@short ;
   talk:inverse "has $o in it"@default ,
               "includes"@short ,
               "includes $o in its cast"@long ;
   rdfs:label "ACTED_IN" .

mv:WROTE rdf:type owl:ObjectProperty ;
   rdfs:domain mv:Person ; rdfs:range mv:Movie ;
   talk:direct "is the author of"@default ,
               "wrote"@short ,
               "wrote the script of"@long ;
   talk:inverse "is authored by"@default ,
               "is written by"@long , "is by"@short ;
   rdfs:label "WROTE" .

mv:born rdf:type owl:DatatypeProperty ;
   rdfs:domain mv:Person ;
   talk:direct "The birth year of $s was"@long ,
               "spawned in"@short ,
               "was born in"@default ;
   rdfs:label "born" .

mv:released rdf:type owl:DatatypeProperty ;
   rdfs:domain mv:Movie ;
   talk:direct "premiered in"@short ,
               "The release year of $s was"@long ,
               "was released in"@default ;
   rdfs:label "released" .

mv:tagline rdf:type owl:DatatypeProperty ;
   rdfs:domain mv:Movie ;
   talk:direct "'s tagline goes: '$o'"@default ,
               "'s tagline is"@short ,
               "the tagline for $s was"@long ;
   rdfs:label "tagline" .

MATCH (lemma:Form)<-[:canonicalForm]-(le:LexicalEntry)
  -[:sense]->()-[:isLexicalizedSenseOf]->(concept)
WHERE lemma.writtenRep = "clear"
RETURN le.partOfSpeech AS PoS,
       concept.definition AS definition

// clear는 명사·동사·형용사·부사 노릇을 다 하니
// 결과가 수두룩하다 — WordNet 공개 웹 검색과
// 정확히 일치한다(그림 13-5). RETURN path로
// 바꾸면 그림 13-6의 시각화가 된다

MATCH (a:LexicalConcept {uri: $a_id})
MATCH (b:LexicalConcept {uri: $b_id })
MATCH p = shortestPath((a)-[:hypernym*0..]-(b))
WITH a, b, length(p) AS pathLen
RETURN a.definition AS a_def, b.definition AS b_def,
  pathLen, 1.0/(1+pathLen) AS pathSim

// :params
// a_id: "https://en-word.net/id/oewn-02086723-n" (dog)
// b_id: "https://en-word.net/id/oewn-02131817-n" (lion)

from nltk.corpus import wordnet as wn

dog = wn.synset('dog.n.01')
lion = wn.synset('lion.n.01')
print(dog.definition())
print(lion.definition())
print(dog.path_similarity(lion))

# a member of the genus Canis (probably descended
#   from the common wolf) that has been domesticated
#   by man since prehistoric times; occurs in many breeds
# large gregarious predatory feline of Africa and India
#   having a tawny coat with a shaggy mane in the male
# 0.16666666666666666   ← Cypher와 정확히 같다 (표 13-2)

MATCH path = (leaf:LexicalConcept)
    -[:hypernym*0..]->(root)
WHERE NOT EXISTS (()-[:hypernym]->(leaf))
  AND NOT EXISTS ((root)-[:hypernym]->())
RETURN max(length(path)) AS maxTaxonomyDepth

// 이 그래프의 전역 최대치는 19.
// 주의 — 모든 범주가 단일 루트에 이어져 있으란 법은
// 없다. WordNet도 다중 루트, 즉 서로 끊어진 여러
// 택소노미를 가진 그래프다. 그럴 때는 비교 대상
// 요소들을 담은 택소노미의 깊이를 계산해야 한다

MATCH (a:LexicalConcept {uri: $a_id})
MATCH (b:LexicalConcept {uri: $b_id })
MATCH p = (a)-[:hypernym*0..]->(lcs)
          <-[:hypernym*0..]-(b)
WITH a, b, lcs, length(p) AS pathLen
MATCH p = (lcs)-[:hypernym*0..]->(root)
  WHERE NOT (root)-[:hypernym]->()
RETURN lcs.definition AS lcs_def, lcs.uri AS lcs_id,
  length(p) AS lcs_depth
ORDER BY pathLen LIMIT 1

// 표 13-3의 결과 —
// lcs_def: "a terrestrial or aquatic
//           flesh-eating mammal"
// lcs_id:  …/id/oewn-02077948-n
// lcs_depth: 11

MATCH (a:LexicalConcept {uri: $a_id})
MATCH (b:LexicalConcept {uri: $b_id })
MATCH (lcs:LexicalConcept {uri: $lcs_id })
MATCH a_to_lcs = (a)-[:hypernym*0..]->(lcs)
MATCH b_to_lcs = (b)-[:hypernym*0..]->(lcs)
WITH a, b, lcs, length(a_to_lcs) AS depth_a,
  length(b_to_lcs) AS depth_b
RETURN (2.0 * $lcs_depth) /
  (2.0 * $lcs_depth + depth_a + depth_b)
  AS wp_sim

# NLTK — 함수 이름만 다르다:
# print(wn.wup_similarity(dog, lion))

MATCH (prop:LexicalConcept {
  uri:"https://custom.extension/id/15349-n"})
MATCH (oss:LexicalConcept {
  uri:"https://custom.extension/id/15350-n"})
MATCH (sw:LexicalConcept {
  uri:"https://en-word.net/id/oewn-06578068-n"})
MERGE (prop)-[:hypernym]->(sw)-[:hyponym]->(prop)
MERGE (oss)-[:hypernym]->(sw)-[:hyponym]->(oss)

// software(oewn-06578068-n)에 hypernym과
// 대칭 hyponym으로 갈고리를 건다.
// Example 13-40의 탐색 질의로 새 부분 그래프를
// 시각 확인할 수 있다 (그림 13-14)