I needed textual information about the entities in knowledge completion datasets so I aquired it. I'm sharing it here, no proof for correctness. Use it with caution.
Under other/
you can find other (mostly toyish) KGC datasets where no text matching has been done.
These datasets are based on the Freebase Knowledge Graph and entities are mentioned
by their Freebase id. As the Freebase KG is archived and not in use anymore, I matched
the entities with Wikidata entities and obtained metadata from Wikidata. Wikidata entities
contain a freebase_id
relation, which was used to match the entities. However, not all
entities could be resolved that way so I queried DBPedia for the remaining.
There still remained about ~40 entities for which no textual information could be found.
See the entity2wikidata.json file for metadata about the Freebase entities.
def freebase2wikidata(entities):
"""
This method constructs a dictionary mapping an freebase id to some wikidata entities.
:param entities: an iterable of string entities
:return:
"""
import requests
from SPARQLWrapper import SPARQLWrapper, JSON
sparql = SPARQLWrapper("http://dbpedia.org/sparql")
sparql.setReturnFormat(JSON)
def dbpedia_with_freebase(entities):
"""
:param entities: list of entities
:return: dict: { "freebase" : { "wikidata1" : {},
"wikidata2" : {},
},
...}
"""
### Part 1 ####
# Query DBPedia for Wikidata Ids
# finds all wikidata_ids that have this freebase id
dbpedia_query = """PREFIX dbpedia: <http://dbpedia.org/resource/>
SELECT DISTINCT ?other WHERE {
?obj (owl:sameAs) <http://rdf.freebase.com/ns/%s>.
?obj (owl:sameAs) ?other .
FILTER (strstarts(str(?other), 'http://www.wikidata.org/entity/'))
}"""
res = {}
for e in entities:
q = dbpedia_query % e[1:].replace('/', '.') # /m/xxxx -> m.xxxx
sparql.setQuery(q)
results = sparql.query().convert()
for result in results["results"]["bindings"]:
if e not in res:
res[e] = {}
wd = result['other']['value'].replace(
'http://www.wikidata.org/entity/', '')
res[e][wd] = {}
return res
def wikidata_with_freebase(entities):
"""
:param entities: list of freebase entities
:return: dict {
'/m/01bs9f': {'Q13582652': {'alternatives': set(),
'description': 'engineer specialising
in design, construction
and maintenance of the
built environment',
'label': 'civil engineer',
'wikipedia': set()
}
},
'/m/01cky2': ...
}
"""
query_wikidata_with_freebase = '''
PREFIX wikibase: <http://wikiba.se/ontology#>
PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT DISTINCT ?wd ?fb ?wdLabel ?wdDescription ?alternative ?sitelink
WHERE {
?wd wdt:P646 ?fb .
OPTIONAL { ?wd schema:description ?itemdesc . }
OPTIONAL { ?wd skos:altLabel ?alternative .
FILTER (lang(?alternative) = "en").
}
OPTIONAL { ?sitelink schema:about ?wd .
?sitelink schema:inLanguage "en" .
FILTER (SUBSTR(str(?sitelink), 1, 25) = "https://en.wikipedia.org/") .
} .
VALUES ?fb { "%s" }
SERVICE wikibase:label { bd:serviceParam wikibase:language "en". }
}'''
url = 'https://query.wikidata.org/bigdata/namespace/wdq/sparql'
res = {}
for ents in zip(*(iter(entities),) * 100):
query_ = query_wikidata_with_freebase % '" "'.join(ents)
data = requests.get(url, params={'query': query_, 'format': 'json'}).json()
for item in data['results']['bindings']:
wd = item['wd']['value'].replace('http://www.wikidata.org/entity/', '')
fb = item['fb']['value']
label = item['wdLabel']['value'] if 'wdLabel' in item else None
desc = item['wdDescription']['value'] if 'wdDescription' in item else None
alias = {item['alternative']['value']} if 'alternative' in item else set()
sitelink = {item['sitelink']['value']} if 'sitelink' in item else set()
if fb not in res:
res[fb] = {}
if wd not in res[fb]:
res[fb][wd] = {'label': label,
'description': desc,
'wikipedia': sitelink,
'alternatives': alias}
res[fb][wd]['wikipedia'] |= sitelink
res[fb][wd]['alternatives'] |= alias
return res
def wikidata_with_wikidata(entities):
"""
:param dict entities: { "freebase" : { "wikidata1" : {},
"wikidata2" : {},
},
...}
:return: dict {
'/m/01bs9f': {'Q13582652': {'alternatives': set(),
'description': 'engineer specialising
in design, construction
and maintenance of the
built environment',
'label': 'civil engineer',
'wikipedia': set()
}
},
'/m/01cky2': ...
}
"""
query_wd_with_wd = '''PREFIX wikibase: <http://wikiba.se/ontology#>
PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT DISTINCT ?wd ?fb ?wdLabel ?wdDescription ?alternative ?sitelink
WHERE {
BIND(wd:%s AS ?wd).
OPTIONAL { ?wd schema:description ?itemdesc . }
OPTIONAL { ?wd skos:altLabel ?alternative .
FILTER (lang(?alternative) = "en").
}
OPTIONAL { ?sitelink schema:about ?wd .
?sitelink schema:inLanguage "en" .
FILTER (SUBSTR(str(?sitelink), 1, 25) = "https://en.wikipedia.org/") .
} .
SERVICE wikibase:label { bd:serviceParam wikibase:language "en". }
}'''
url = 'https://query.wikidata.org/bigdata/namespace/wdq/sparql'
res = {}
for fb, wd_ids in entities.items():
for wd_id in wd_ids:
query_ = query_wd_with_wd % wd_id
data = requests.get(url,
params={'query': query_, 'format': 'json'}).json()
for item in data['results']['bindings']:
wd = item['wd']['value'].replace('http://www.wikidata.org/entity/',
'')
# fb = item['fb']['value']
label = item['wdLabel']['value'] if 'wdLabel' in item else None
desc = item['wdDescription'][
'value'] if 'wdDescription' in item else None
alias = {
item['alternative']['value']} if 'alternative' in item else set()
sitelink = {
item['sitelink']['value']} if 'sitelink' in item else set()
if fb not in res:
res[fb] = {}
if wd not in res[fb]:
res[fb][wd] = {'label': label,
'description': desc,
'wikipedia': sitelink,
'alternatives': alias}
res[fb][wd]['wikipedia'] |= sitelink
res[fb][wd]['alternatives'] |= alias
return res
# lets first try to find the freebase entities in wikidata
result = wikidata_with_freebase(entities)
logging.info("Found %s freebase entities in wikidata (from total %s)." %
(len(result), len(entities)))
# then find the remaining ids in dbpedia
missing_entities = set(entities) - set(result.keys())
result_missing = dbpedia_with_freebase(missing_entities)
# and query the wikidata information afterwards
result_missing = wikidata_with_wikidata(result_missing)
logging.info("Found %s missing entities via dbpedia in wikidata (from total %s "
"missing entities)." %
(len(result_missing), len(missing_entities)))
# merge the two dicts
result = {**result, **result_missing}
# and remove the sets
for fb, wds in result.items():
for wd_id, stats in wds.items():
result[fb][wd_id]['wikipedia'] = stats['wikipedia'].pop() if stats[
'wikipedia'] else None
result[fb][wd_id]['alternatives'] = list(stats['alternatives'])
logging.info("Final: Found %s freebase entities in wikidata (from total %s)." %
(len(result), len(entities)))
return result
I wanted to work with the datasets WN18 and WN18RR that contain 18/11 relations from wordnet data.
The original WN18RR dataset has the following form:
02174461 _hypernym 02176268
05074057 _derivationally_related_form 02310895
08390511 _synset_domain_topic_of 08199025
02045024 _member_meronym 02046321
01257145 _derivationally_related_form 07488875
...
I wanted to have the textual representation of the entities, but only the wordnet offsets are given as entites, transforming them back is problematic cause they are ambiguous within the 4 datafiles from wordnet.
For example 01257145 _derivationally_related_form 07488875
has two offsets:
01257145
and 07488875
.
01257145 | 07488875 | |
---|---|---|
ADJ | sensual.s.02 |
|
ADV | ||
NOUN | precession.n.02 |
sensuality.n.01 |
VERB |
I transformed the dataset back to wordnet synsets by validating if the given relation holds between the ambiguous entities.
The transformed textual data then looks like this:
clangor.v.01 _hypernym sound.v.02
straightness.n.02 _derivationally_related_form straight.a.02
militia.n.01 _synset_domain_topic_of military.n.01
alcidae.n.01 _member_meronym pinguinus.n.01
sensual.s.02 _derivationally_related_form sensuality.n.01
You can load it into NLTK by executing
from nltk.corpus import wordnet as wn
wn.synset('sensual.s.02')
As first stated by Toutanova in 2015 and confirmed by Dettmers in 2018, the dataset suffers from informative value, cause >80% of the test triples (e1, r1, e2) can be found in the training set with another relation: (e1, r2, e2) or (e2, r2, e1). Dettmers used a rule-based model which learned the inverse relation and achieved state-of-the-art results on that dataset. It should therefore not used for research evaluation anymore.
I got the WN18RR dataset from TimDettmers/ConvE. As the original WN18 is down, I obtained a copy from Github.