def get_top_ppr(ppr_seed, max_num=10, lang="en"):
'''
Get nodes based on Personalized PageRank(PPR).
:param ppr_seed: PPR seed node ID.
:param max_num: maximum number of records to retrieve.
:param lang: language to return
:return: list type, list of the records output by Neo4j query
'''
if lang == "en":
lang = ""
else:
lang = "_" + lang
results = []
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
tx = session.begin_transaction()
ppr_result = tx.run(
f"MATCH (p) WHERE ID(p) = {ppr_seed} "
"WITH p CALL gds.pageRank.stream('memex-graph', { maxIterations: 10, dampingFactor: 0.85, sourceNodes: [p] }) YIELD nodeId, score AS pageRank "
"RETURN gds.util.asNode(nodeId).wid, gds.util.asNode(nodeId).label"
+ lang + ", gds.util.asNode(nodeId).description" + lang +
", gds.util.asNode(nodeId).guid "
"ORDER BY pageRank DESC "
f"LIMIT {max_num}")
for record in ppr_result:
results.append(record)
return results
def get_node_details(node_id, lang="en"):
'''
Get node main details.
:param node_id: the ID of the source node to get its main details.
:param lang: language to return
:return: node main details
'''
if lang == "en":
lang = ""
else:
lang = "_" + lang
results = []
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
tx = session.begin_transaction()
result = tx.run("MATCH (p) "
f"WHERE p.wid='{node_id}' "
"RETURN p.wid, p.label" + lang + ", p.description" +
lang + ", p.guid, p.image, p.location.x, p.location.y")
for record in result:
results.append(record)
if results:
return results[0]
else:
return db_conn._error_status
def get_top_closest(tx, n, long, lat, meters, lang="en"):
'''
Search for entites using location information.
:param tx: transaction object reference
:param n: integer type, number of records
:param long: float type, longitude coordinates
:param lat: float type, latitude coordinates
:param meters: integer type, radial distance from point in meters
:param lang: language to return
:return: list type, list of the records output by Neo4j query
'''
if lang == "en":
lang = ""
else:
lang = "_" + lang
results = []
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
tx = session.begin_transaction()
result = tx.run(
"MATCH (p) "
"WHERE distance(point({latitude:$lat, longitude:$long}),p.location) < $meters "
"RETURN p.label_" + lang +
",p.wid,distance(point({latitude:$lat, longitude:$long}),p.location) AS dist, "
" id(p) "
"ORDER BY dist ASC LIMIT $n",
lat=lat,
long=long,
meters=meters,
n=n)
for record in result:
results.append(record)
return results
def get_top_similar_closest(search_mode, sentence, n, long, lat, meters):
'''
Search for entites using text and location information based on the different search modes (semantic embeddings or fulltext).
:param search_mode: str type, search mode semantic or fulltext (exact search)
:param sentence: str type, textual description
:param n: integer type, subset the top ranked results
:param long: float type, longitude coordinates
:param lat: float type, latitude coordinates
:param meters: integer type, radial distance from point in meters
:return: list of top n closest points by distance and similarity
'''
db_conn = db_connection.DB_Connection()
if (search_mode == "semantic"):
prova = [float(x) for x in get_emb_vect(sentence)]
with db_conn._driver.session() as session:
results = session.read_transaction(get_within_radius_w2v,
long=long,
lat=lat,
meters=meters,
query=prova)
else:
with db_conn._driver.session() as session:
results = session.read_transaction(get_within_radius_fulltext,
long=long,
lat=lat,
meters=meters,
query=sentence)
return results[:n]
def get_node_neighbors(node_id, lang="en"):
'''
Get the 1-hop neighbors of a node and their relation to the source node.
:param node_id: the ID of the source node to get its neighbors.
:param lang: language to return
:return: list of relation type, neighbor id and neighbor label
'''
if lang == "en":
lang_rel = ".en"
lang = ""
else:
lang_rel = "." + lang
lang = "_" + lang
results = []
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
tx = session.begin_transaction()
result = tx.run("MATCH (n)-[r]->(b) "
f"WHERE n.wid='{node_id}' "
"RETURN r" + lang_rel + ", b.wid, b.label" + lang)
for record in result:
results.append(record)
return results
def get_top_visually_similar_closest(image_emb,
n,
long,
lat,
meters,
lang="en"):
'''
Search for entites using image and location information.
:param image_emb: float array
:param n: integer type, subset the top ranked results
:param long: float type, longitude coordinates
:param lat: float type, latitude coordinates
:param meters: integer type, radial distance from point in meters
:param lang: language to return
:return: list of top n closest points by distance and similarity
'''
results = []
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
results = session.read_transaction(get_within_radius_visual,
long=long,
lat=lat,
meters=meters,
query=image_emb,
lang=lang)
return results[:n]
def update_fulltext_index(lang="en"):
'''
Drop the search index if exists and create a new one over the entity labels and descriptions.
:param lang: language to return
'''
if lang == "en":
lang = ""
else:
lang = "_" + lang
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
try:
tx = session.begin_transaction()
tx.run("CALL db.index.fulltext.drop('labelsAndDescriptions')")
tx.commit()
except:
pass
with db_conn._driver.session() as session:
try:
tx = session.begin_transaction()
tx.run(
"CALL db.index.fulltext.createNodeIndex('labelsAndDescriptions', ['Knowledge', 'Place'], ['label"
+ lang + "', 'description" + lang + "'])")
tx.commit()
except:
pass
def europeana_ingest_pilots(cities,
link_to_nodes=False,
annotation_threshold=0.1):
"""
Reads the list of desired cities to ingest (Default: 'list_cities.txt')
Retrieves the list of europeana items and adds to Neo4j.
:param cities: list of cities to retrieve
:param link_to_nodes: a flag for linking the ingested entities to existing entities in the graph
:param annotation_threshold: TagMe annotation threshold
"""
db_conn = db.DB_Connection()
wikidata_found = False
for city in cities:
print("Europeana ingestion...", city)
data = eip.query_only_id_bboxes(
city,
db_conn=db_conn,
link_to_nodes=link_to_nodes,
annotation_threshold=annotation_threshold)
# print("Processing ", len(data), " entities...")
# for n in tqdm(data):
# if link_to_wikidata:
# #Add wikidata wids
# wiki_titles = []
# for idx, property_name in enumerate(n[0]):
# if property_name == "label" or property_name == "description" or property_name == "dcCreator":
# value = n[1][idx]
# europeana_annotations = tagme.annotate(value)
# if europeana_annotations:
# for ann in europeana_annotations.get_annotations(annotation_threshold):
# t = ann.entity_title
# t = t[0].lower() + t[1:]
# wiki_titles.append(t)
#
# if property_name == "dcCreator":
# print(n[1][idx])
#
# wiki_titles = list(dict.fromkeys(wiki_titles))
# all_wids = []
# for title in wiki_titles:
# wids = db_conn.get_wikidata_ids_by_label(title)
# if wids:
# all_wids.extend(wids)
#
# if all_wids:
# wikidata_found = True
# n[0].append("wids")
# n[1].append(all_wids)
#
# #Insert the node
# db_conn.queue_insert_node(n)
#
# #Link to wikidata nodes
# if link_to_wikidata and wikidata_found:
# db_conn.match_with_wikidata()
db_conn.close()
def mapillary_ingest_pilots(cities):
"""
Reads the list of desired cities to ingest (Default: 'list_cities.txt')
Retrieves the list of mapillary image items and adds to Neo4j.
:param cities: list of cities to retrieve
"""
db_conn = db.DB_Connection()
for city in cities:
print("Mapillary ingestion...", city)
data = mip.query_only_id_bboxes(city)
for n in data:
db_conn.queue_insert_node(n)
db_conn.close()
def get_top_visually_similar(image_emb, n, lang="en"):
'''
Search for entites using image information.
:param image_emb: float array
:param n: integer type, subset the top ranked results
:param lang: language to return
:return: list of top n closest points and visual similarity
'''
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
results = session.read_transaction(get_visual,
query=image_emb,
lang=lang)
return results[:n]
def delete_story(story_id):
"""
Delete story entity from the KG.
:return: JSON response with the success status True/False
"""
db_conn = db.DB_Connection()
result = db_conn.get_story(story_id)
if result == -1:
return jsonify(success=False), 500
else:
status = db_conn.delete_story(story_id)
if status == 0:
return jsonify(success=True)
else:
return jsonify(success=False), 500
def get_stories_by_location():
"""
Retrieves stories by a given location.
:return: JSON response with the list of stories
"""
req_data = request.get_json()
if 'latitude' in req_data:
latitude = int(req_data["latitude"])
else:
return jsonify({'message': 'Missing latitude field.'}), 400
if 'longitude' in req_data:
longitude = int(req_data["longitude"])
else:
return jsonify({'message': 'Missing longitude field.'}), 400
if 'meters' in req_data:
meters = int(req_data["meters"])
else:
return jsonify({'message': 'Missing meters field.'}), 400
if 'max_num' in req_data:
max_num = int(req_data["max_num"])
else:
max_num = 10
db_conn = db.DB_Connection()
result = db_conn.get_stories_by_location(latitude, longitude, meters,
max_num)
res = {}
for i in range(len(result)):
temp = {}
temp['story_id'] = result[i][0]
temp['label'] = result[i][1]
temp['description'] = result[i][2]
temp['image'] = result[i][3]
temp['audio'] = result[i][4]
temp['video'] = result[i][5]
temp['latitude'] = result[i][6]
temp['longitude'] = result[i][7]
temp['dist'] = result[i][8]
res[i] = temp
return res
def get_wids(overwrite=True, with_image=False):
'''
Get the wids and labels of all the entites in the KG.
:param overwrite: flag if true return entity wids regardless of having descriptions or not
:param with_image: filters for only nodes that have images
:return: list of entity wids and labels
'''
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
if overwrite:
if with_image:
query = """
MATCH (p) WHERE EXISTS (p.wid) AND EXISTS (p.image)
RETURN p.wid, p.label, p.image
"""
else:
query = """
MATCH (p) WHERE EXISTS (p.wid)
RETURN p.wid, p.label
"""
else:
if with_image:
query = """
MATCH (p) WHERE EXISTS (p.wid) AND EXISTS (p.image) AND (NOT EXISTS (p.description) OR NOT EXISTS (p.image_emb))
RETURN p.wid, p.label, p.image
"""
else:
query = """
MATCH (p) WHERE EXISTS (p.wid) AND NOT EXISTS (p.description)
RETURN p.wid, p.label
"""
results = session.run(query)
l = []
if with_image:
for i in results:
l.append([i[0], i[1], i[2]])
else:
for i in results:
l.append([i[0], i[1]])
db_conn.close()
return l
def update_desc_emb(nodes, lang='en', wikipedia_sentences=2):
'''
Get node descriptions from Wikidata enriched with Wikipedia descriptions,
generate embeddings based on the labels and descriptions,
and set the embeddings as node property.
:param nodes: dict, node ids and labels
:param lang: language (default 'en')
:param wikipedia_sentences: number of sentences to get from wikipedia (default 2)
:return: a confirmation message
'''
db_conn = db_connection.DB_Connection()
session = db_conn._driver.session()
print(len(nodes))
for wid, label in tqdm(nodes):
wikipedia_descr = ""
try:
wikipedia_descr = wikipedia.summary(label,
sentences=wikipedia_sentences)
except:
pass
desc = ""
try:
entity = client.get(wid, load=True)
if entity.description:
descriptions = entity.description.texts
if bool(descriptions):
if lang in descriptions.keys():
desc = descriptions[lang]
except:
pass
text = label + "\n" + desc + "\n" + wikipedia_descr
emb = avg_feature_vector(text)
emb = [float(x) for x in emb]
set_embedding_w2v(session, wid, desc, emb, wikipedia_descr)
return "Desc and embeddings updated successfully"
def wikidata_ingest_pilots(hops, cities):
"""
Reads the list of desired cities to ingest (Default: 'list_cities.txt')
Retrieves the list of wikidata entity ids for each city, and calls
download_n_hops_rec function in order to start recursive retrieval.
:param hops: number of recursive hops
:param cities: list of cities to retrieve
"""
db_conn = db.DB_Connection()
for city in cities:
print("Wikidata ingestion...", city)
data = wip.query_only_id_bboxes(city)
starting_urls = []
for line in data:
# line is a wid of the format "wd:Q* ", we need only "Q* "
starting_urls.append(line['place'].split("/")[-1])
rhi.download_n_hops_rec(starting_urls, db_conn, hops)
db_conn.close()
def get_top_similar(search_mode, sentence, n):
'''
Search for entites using text based on the different search modes (semantic embeddings or fulltext).
:param search_mode: str type, search mode
:param sentence: str type, textual description
:param n: integer type, subset the top ranked results
:return: list of top n closest points by distance and similarity
'''
db_conn = db_connection.DB_Connection()
if (search_mode == "semantic"):
prova = [float(x) for x in get_emb_vect(sentence)]
with db_conn._driver.session() as session:
results = session.read_transaction(get_sim_w2v, query=prova)
else:
with db_conn._driver.session() as session:
results = session.read_transaction(get_sim_fulltext,
query=sentence)
return results[:n]
def get_story(story_id):
"""
Retrieves story details based on a given story ID.
:return: JSON response with the story details
"""
db_conn = db.DB_Connection()
result = db_conn.get_story(story_id)
if result == -1:
return jsonify(success=False), 500
else:
resp = {}
resp['label'] = result[0]
resp['description'] = result[1]
resp['image'] = result[2]
resp['audio'] = result[3]
resp['video'] = result[4]
resp['latitude'] = result[5]
resp['longitude'] = result[6]
return resp
def update_named_graph():
'''
Drop the named graph if exists and create a new one over "Place" and "Knowledge" nodes.
'''
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
try:
tx = session.begin_transaction()
tx.run("CALL gds.graph.drop('memex-graph')")
tx.commit()
except:
pass
with db_conn._driver.session() as session:
try:
tx = session.begin_transaction()
tx.run(
"CALL gds.graph.create('memex-graph',['Place', 'Knowledge'],'*')"
)
tx.commit()
except:
pass
def get_similar_stories():
"""
Retrieves stories similar to a given story.
:return: JSON response with the list of stories
"""
req_data = request.get_json()
if 'story_id' in req_data:
story_id = str(req_data["story_id"])
story_id = "'" + story_id + "'"
else:
return jsonify({'message': 'Missing story_id field.'}), 400
db_conn = db.DB_Connection()
result = db_conn.get_story(story_id)
if result == -1:
return jsonify(success=False), 500
else:
if 'max_num' in req_data:
max_num = int(req_data["max_num"])
else:
max_num = 10
result = db_conn.get_similar_stories(story_id, max_num)
resp = {}
for i in range(len(result)):
temp = {}
temp['story_id'] = result[i][0]
temp['label'] = result[i][1]
temp['description'] = result[i][2]
temp['image'] = result[i][3]
temp['audio'] = result[i][4]
temp['video'] = result[i][5]
temp['latitude'] = result[i][6]
temp['longitude'] = result[i][7]
temp['sim'] = result[i][8]
resp[i] = temp
return resp
def update_visual_desc_emb(nodes,
model='ResNet18',
temp_folder='data/images/',
keep_images=False,
throttle=2):
'''
Fetches the image creates a visual embedding using the described model
:param nodes: array of arrays [[node ids, label, image url]]
:temp_folder: the folder the images will be downloaded (temporarily if keep_images = false)
:keep_images: whether to not delete the image
:return: True on success
'''
from db.models import prepare_image_from_file, get_visual_model, get_visal_embedding
import time
os.makedirs(temp_folder, exist_ok=True)
db_conn = db_connection.DB_Connection()
session = db_conn._driver.session()
model = get_visual_model(model)
for wid, label, image_url in tqdm(nodes):
# Encode only the first image (most common scenario)
if type(image_url) == list:
image_url = image_url[0]
ext = image_url[-3:]
filepath = temp_folder + wid + '.' + ext
filename = wget.download(image_url, filepath)
I = prepare_image_from_file(filepath)
if type(I) == type(None):
continue
emb = get_visal_embedding(model, I)
emb = [float(x) for x in emb]
set_embedding_visual(session, wid, emb)
if not keep_images:
os.remove(filepath)
time.sleep(throttle)
return True
def get_ppr_subgraph(ppr_seed, max_num=10):
'''
Get a PPR based subgraph.
:param ppr_seed: PPR seed node ID.
:param max_num: maximum number of records to retrieve.
:return: list type, list of the records output by Neo4j query
'''
db_conn = db_connection.DB_Connection()
with db_conn._driver.session() as session:
tx = session.begin_transaction()
ppr_result = tx.run(
f"MATCH (p) WHERE ID(p) = {ppr_seed} "
"WITH p CALL gds.pageRank.stream('memex-graph', { maxIterations: 10, dampingFactor: 0.85, sourceNodes: [p] }) YIELD nodeId, score AS pageRank "
f"WITH gds.util.asNode(nodeId) AS n, pageRank "
"MATCH (n)-[r]-(b) "
"WHERE EXISTS(b.wid) "
"RETURN * "
"ORDER BY pageRank DESC "
f"LIMIT {max_num}+10")
ppr_result.graph()
return ppr_result
def get_story_connections():
"""
Retrieves the KG entities connected to a given story.
:return: JSON response with the list of KG entities
"""
req_data = request.get_json()
if 'story_id' in req_data:
story_id = str(req_data["story_id"])
story_id = "'" + story_id + "'"
else:
return jsonify({'message': 'Missing story_id field.'}), 400
if 'language' in req_data:
lang = str(req_data["language"])
else:
lang = "en"
db_conn = db.DB_Connection()
result = db_conn.get_story(story_id)
if result == -1:
return jsonify(success=False), 500
else:
result = db_conn.get_story_connections(story_id, lang)
resp = {}
for i in range(len(result)):
temp = {}
temp['wid'] = result[i][0]
temp['label'] = result[i][1]
temp['description'] = result[i][2]
temp['image'] = result[i][3]
temp['latitude'] = result[i][4]
temp['longitude'] = result[i][5]
resp[i] = temp
return resp
def add_story():
"""
Add story entity to the KG.
:return: JSON response with the success status True/False
"""
req_data = request.get_json()
properties = []
values = []
if 'story_id' in req_data:
story_id = str(req_data["story_id"])
properties.append('story_id')
values.append(story_id)
else:
return jsonify({'message': 'Missing story_id field.'}), 400
if 'label' in req_data:
properties.append('label')
values.append(req_data['label'])
if 'description' in req_data:
properties.append('description')
values.append(req_data['description'])
if 'keywords' in req_data:
properties.append('keywords')
values.append(req_data['keywords'] + [' '])
if 'image' in req_data:
properties.append('image')
values.append(req_data['image'])
if 'audio' in req_data:
properties.append('audio')
values.append(req_data['audio'])
if 'video' in req_data:
properties.append('video')
values.append(req_data['video'])
if 'latitude' and 'longitude' in req_data:
properties.append('coordinate_location')
values.append([req_data['latitude'], req_data['longitude']])
data = [properties, values]
db_conn = db.DB_Connection()
result = db_conn.add_story(data)
if result == 0:
if 'language' in req_data:
lang = str(req_data["language"])
else:
lang = "en"
db_conn.connect_story_by_textual_sim("'" + story_id + "'", "Knowledge",
lang)
db_conn.connect_story_by_textual_sim("'" + story_id + "'", "Place",
lang)
if 'connections' in req_data:
connections = req_data['connections']
db_conn.connect_story_to_target_nodes("'" + story_id + "'",
connections)
db_conn.link_stories_together()
return jsonify(success=True)
else:
return jsonify(success=False)
def edit_story(story_id):
"""
Edit story entity in the KG.
:return: JSON response with the success status True/False
"""
req_data = request.get_json()
properties = []
values = []
if 'label' in req_data:
properties.append('label')
values.append(req_data['label'])
if 'description' in req_data:
properties.append('description')
values.append(req_data['description'])
if 'keywords' in req_data:
properties.append('keywords')
values.append(req_data['keywords'] + [' '])
if 'image' in req_data:
properties.append('image')
values.append(req_data['image'])
if 'audio' in req_data:
properties.append('audio')
values.append(req_data['audio'])
if 'video' in req_data:
properties.append('video')
values.append(req_data['video'])
if 'latitude' and 'longitude' in req_data:
properties.append('coordinate_location')
values.append([req_data['latitude'], req_data['longitude']])
data = [properties, values]
db_conn = db.DB_Connection()
result = db_conn.get_story(story_id)
if result == -1:
return jsonify(success=False), 500
else:
result = db_conn.edit_story(str(story_id), data)
db_conn.remove_story_connections(str(story_id))
if 'language' in req_data:
lang = str(req_data["language"])
else:
lang = "en"
db_conn.connect_story_by_textual_sim(str(story_id), "Knowledge", lang)
db_conn.connect_story_by_textual_sim(str(story_id), "Place", lang)
if 'connections' in req_data:
connections = req_data['connections']
db_conn.connect_story_to_target_nodes("'" + story_id + "'",
connections)
db_conn.link_stories_together()
if result == 0:
return jsonify(success=True)
else:
return jsonify(success=False), 500
type=int,
help='[Mode 8] Get the top K')
parser.add_argument('--query',
metavar='text query',
type=str,
help='[Mode 8] Text query')
parser.add_argument('--meters',
default=5000,
metavar='text query',
type=int,
help='[Mode 8] Radial distance in meters')
parser.add_argument('--verbose', type=str2bool, default=False)
args = parser.parse_args()
# Check if Community
db_conn = db.DB_Connection()
neo4j_version = db_conn.version()
db_conn.close()
if args.verbose:
print("[INFO] Neo4j Currently running on", neo4j_version)
#if 'community' not in neo4j_version:
#print('not community')
#db_conn = db.DB_Connection()
#db_conn.initalise_neo4j()
#db_conn.close()
if args.mode == 0: # Wikidata ingestion tool
if args.cities:
f = open(args.cities, 'r')
cities = []
