-
Notifications
You must be signed in to change notification settings - Fork 0
/
colab_gemini.py
924 lines (690 loc) · 776 KB
/
colab_gemini.py
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
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
# -*- coding: utf-8 -*-
"""Gemini.ipynb
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1-ebU3UMTEEK21HfflGEwQWnQgmAu8vix
# Gemini 👥
An eclectic news search engine that helps with deduplication of news articles and enriches your reading experience. Enables readers to dig deep into a certain opinion or explore multiple facets of the news at hand.
"""
# Necessary installs -- all are open source libraries
!pip install -U PyTigerGraph keybert sentence-transformers newscatcherapi
"""# Using NewsCatcher API to collect news dataset
Use double quotes (") for the exact match.
If you write *q=Sharan Babu* it will be treated as *q=Sharan AND Babu*
"""
# SET YOUR SEARCH TERM HERE !!!
search_term = "Russia vs Ukraine war"
# Fetch relevant news from our API
from newscatcherapi import NewsCatcherApiClient
API_KEY = '4-MzQydFTLmS_18AZEQYpEVUuUfw4UrmBXI22hayRpM' # Replace with your API key (Website Link: https://app.newscatcherapi.com/auth/register)
newscatcherapi = NewsCatcherApiClient(x_api_key=API_KEY)
news_articles = newscatcherapi.get_search(q=search_term, lang='en', page_size=100)
# View collected news articles
news_articles
"""## Pickle and Unpickle news results
To reduce number of calls to API and Reuse fetched results
"""
# Save fetched data
import pickle
with open('news_results.json','wb') as file:
pickle.dump(news_articles, file)
# load data as and when needed
import pickle
with open('news_results.json','rb') as file:
loaded_news = pickle.load(file)
# View loaded news
loaded_news
"""# Enrich data with useful metadata"""
# Convert News into Pandas Dataframe
# -- or better viewing and processing
import pandas as pd
pandas_table = pd.DataFrame(news_articles['articles'])
# First 5 rows of our news dataset
pandas_table.head()
# Basic look of data -- Contries and number of corresponding articles originating from that country
pandas_table["country"].value_counts()
# List of various topics found in the fetched news (of search term)
pandas_table["topic"].value_counts()
"""# Establish connection to TigerGraph"""
# Connect to our Tigergraph instance
# -- replace values in the cell with your instance details
import pyTigerGraph as tg
import json
import pandas as pd
# Connection parameters
hostName = "https://newssearch.i.tgcloud.io"
userName = "tigergraph"
password = "newssearch19"
conn = tg.TigerGraphConnection(host=hostName, username=userName, password=password)
print("Connected")
"""## DROP ALL DATA / START FRESH"""
# Uncomment below code and Run only if you want to clear all data or start fresh and follow along!!!!!!!!!!
# conn.gsql('''
# USE GLOBAL
# DROP ALL
# ''')
"""## Create Graph Schema"""
# DEFINE / CREATE ALL EDGES AND VERTICES -- Run only once
# Here, we are creating the necessary graph schema according to our use case and required access patterns
results = conn.gsql('''
USE GLOBAL
CREATE VERTEX ANCHOR_NEWS (PRIMARY_ID id STRING, rank INT, summary STRING, link STRING, published_date DATETIME, title STRING, media STRING) WITH primary_id_as_attribute="true"
CREATE VERTEX RELATED_NEWS (PRIMARY_ID id STRING, rank INT, summary STRING, link STRING, published_date DATETIME, title STRING, media STRING) WITH primary_id_as_attribute="true"
CREATE VERTEX KEYWORD (PRIMARY_ID name STRING) WITH primary_id_as_attribute="true"
CREATE VERTEX TOPIC (PRIMARY_ID name STRING) WITH primary_id_as_attribute="true"
CREATE VERTEX COUNTRY (PRIMARY_ID name STRING, positive INT, negative INT)
CREATE UNDIRECTED EDGE similarity_score (From ANCHOR_NEWS, To ANCHOR_NEWS, score FLOAT)
CREATE UNDIRECTED EDGE is_related (From ANCHOR_NEWS, To RELATED_NEWS)
CREATE UNDIRECTED EDGE country_is (From COUNTRY, To RELATED_NEWS | From COUNTRY, To ANCHOR_NEWS)
CREATE UNDIRECTED EDGE topic_is (From TOPIC, To ANCHOR_NEWS | From TOPIC, To RELATED_NEWS)
CREATE UNDIRECTED EDGE keyword_belongs_to (From KEYWORD, To ANCHOR_NEWS)
''')
print(results)
"""
# Preprocess our dataframe
"""
# Drop unnecessary columns
pandas_table.drop(['author', 'published_date_precision','clean_url','excerpt','rights','language','authors','is_opinion','twitter_account','_score'], axis=1, inplace=True)
pandas_table.head(2)
len(pandas_table['title']) # There are a total of 100 rows (news articles) in the pandas dataframe
"""## Semantic Search on summary of news article"""
main_article_summary = pandas_table['summary'][0]
print(main_article_summary)
remaining_articles_array = list(pandas_table['summary'][1:])
remaining_articles_array[0]
semantic_similarity_scores = [1]
# Load Semantic Search Model
from sentence_transformers import SentenceTransformer, util
embedder = SentenceTransformer('msmarco-distilbert-base-tas-b')
main_article_embedding = embedder.encode(main_article_summary)
len(main_article_embedding) # length of intermediate vector produced is 768
remaining_articles_embeddings = embedder.encode(remaining_articles_array)
cos_sim = util.cos_sim(main_article_embedding, remaining_articles_embeddings)
# semantic search example
cos_sim # illustrates similarty between first news article and remaining 99 news articles in our dataset
len(cos_sim[0])
semantic_similarity_scores.extend(cos_sim[0])
len(semantic_similarity_scores)
# Add above semantic scores to our dataframe
# -- This will later be uploaded to our graph and will help us mine useful information
new_array = []
for x in semantic_similarity_scores:
new_array.append(float(x))
print(new_array[0])
# Add new column to pandas table
pandas_table['semantic_score'] = new_array
# view updated dataframe
pandas_table.head(2)
"""## Sentiment Analysis on news Articles"""
import nltk # already installed on Google Colab
from nltk.sentiment import SentimentIntensityAnalyzer
nltk.download('vader_lexicon')
sentiment_model = SentimentIntensityAnalyzer()
# Sentiment analysis example on sample sentence
# -- we wil be applying the same on summaries of news articles and associate them with corresponding countries
sentiment_model.polarity_scores("The russia vs Ukraine war is getting more intense by the hour")
"""## Keyword extraction using NLP"""
# Keyword extraction
from keybert import KeyBERT
doc = """
The United Nations Secretary-General has insisted that the people of Ukraine cannot bear the violence being inflicted on them amid the ongoing war between Russia and Ukraine.
UN Secretary-General, António Guterres, said this while calling for ‘the need to silence the guns and accelerate negotiations towards peace' between Ukraine and Russia.
Guterres also warned that the world is on the brink of a 'perfect storm' as the war in Ukraine exacerbates an already struggling global economy.
In a press release on Wednesday, the UN chief said the war could lead to a 'three-dimensional crisis' in food, energy, and finance, areas that have already been hit hard by COVID-19 and climate change, he said.
"""
kw_model = KeyBERT()
keywords = kw_model.extract_keywords(doc,)
# Example of keywords, extracted from a news article from our dataset
print(keywords)
# Data structures for storing our keywords and sentiment
# -- will later be helpful to easily load data to graph
keyword1 = []
keyword2 = []
keyword3 = []
intent = []
# Takes a while to execute
# -- Create new metadata for our news articles
for x in list(pandas_table['summary']):
keywords = kw_model.extract_keywords(x)
keyword1.append(keywords[0][0])
keyword2.append(keywords[1][0])
keyword3.append(keywords[2][0])
sentiment_result = sentiment_model.polarity_scores(x)
positive = sentiment_result['pos']
negative = sentiment_result['neg']
final_result = ""
if positive > negative:
final_result = "postive"
else:
final_result = "negative"
intent.append(final_result)
# Add these new columns to the pandas table
pandas_table['Keyword 1'] = keyword1
pandas_table['Keyword 2'] = keyword2
pandas_table['Keyword 3'] = keyword3
pandas_table['Intent'] = intent
# View refined dataset
pandas_table.head(2)
"""# Create The Graph and Connect To It"""
# Run only once, after schema is defined
results = conn.gsql('CREATE GRAPH Gemini(ANCHOR_NEWS, RELATED_NEWS, KEYWORD, TOPIC, COUNTRY, similarity_score, is_related, country_is, topic_is, keyword_belongs_to)')
print(results)
# Retrieve Auth token from TigerGraph
conn.graphname="Gemini"
secret = conn.createSecret()
authToken = conn.getToken(secret)
authToken = authToken[0]
print(authToken)
conn = tg.TigerGraphConnection(host=hostName, graphname="Gemini", username=userName, password=password, apiToken=authToken)
def pprint(string):
print(json.dumps(string, indent=2))
"""# Loading Data to Graph"""
# First, let us convert our pandas dataframe to a csv
pandas_table.to_csv('news_data.csv',index=False)
pandas_table.head(1) # View single news article data and meta-data
"""## Terminology:
**Anchor Vertex**: I am calling news articles with unique insights, the ones with different content from the rest as 'Anchor vertices' (*hyper-nodes*). This is because news articles relevant to these (the duplicates) get attached to these anchors according to which anchor they are most similar to.
Anchor vertices are indirectly connected to each other by 'similarity score' edges.
Example image below.
"""
anchor_vertices = []
anchor_vertices.append(pandas_table['summary'][0])
anchor_vertices
# Create mapping between Anchor Vertices and corresponding Children
anchor_to_children = {pandas_table['_id'][0]: []}
anchor_to_children
"""# LOAD DATA"""
# Iterate over dataframe and load data into graph -- Takes a long time (approx 8 to 9 mins for 100 news articles)
countries = {}
keywords = {}
topics = {}
THRESHOLD = 0.7 # 70 percent is the selected similarity threshold, change as per your need
summary_to_id = {}
id_to_semantic = {}
# Populate countries first
for index, row in pandas_table.iterrows():
summary_to_id[row['summary']] = row['_id']
id_to_semantic[row['_id']] = float(row['semantic_score'])
country = row['country']
if country not in countries:
if intent == 'positive':
countries[country] = {'positive':1, 'negative':0}
else:
countries[country] = {'positive':0, 'negative':1}
else:
if intent == 'positive':
countries[country]['positive'] += 1
else:
countries[country]['negative'] += 1
# Create country vertices/upsert country vertices to graph
for each_country in countries:
country_insert_status = conn.upsertVertex("COUNTRY",each_country,{'positive':countries[each_country]['positive'], 'negative':countries[each_country]['negative']})
if country_insert_status == 1:
print(f"Country vertex {each_country} INSERTED successfully")
# Upload all news articles one by one, with their related attributes and edges
for index, row in pandas_table.iterrows():
title = row['title']
published_date = row['published_date']
link = row['link']
summary = row['summary']
rank = int(row['rank'])
topic = row['topic']
country = row['country']
media = row['media']
id = row['_id']
semantic_score = float(row['semantic_score'])
keyword1 = row['Keyword 1']
keyword2 = row['Keyword 2']
keyword3 = row['Keyword 3']
intent = row['Intent']
# Upsert vertices with their data and attributes
if topic not in topics:
topics[topic] = True
topic_insert_status = conn.upsertVertex("TOPIC", topic)
if topic_insert_status == 1:
print(f"Topic vertex {topic} INSERTED successfully")
if keyword1 not in keywords:
keywords[keyword1] = True
keyword1_insert_status = conn.upsertVertex("KEYWORD", keyword1)
if keyword1_insert_status == 1:
print(f"Keyword vertex {keyword1} INSERTED successfully")
if keyword2 not in keywords:
keywords[keyword2] = True
keyword2_insert_status = conn.upsertVertex("KEYWORD", keyword2)
if keyword2_insert_status == 1:
print(f"Keyword vertex {keyword2} INSERTED successfully")
if keyword3 not in keywords:
keywords[keyword3] = True
keyword3_insert_status = conn.upsertVertex("KEYWORD", keyword3)
if keyword3_insert_status == 1:
print(f"Keyword vertex {keyword3} INSERTED successfully")
if index == 0:
# Insert first Anchor article
anchor_insert_status = conn.upsertVertex("ANCHOR_NEWS",id,{'rank':rank,'summary':summary,'link':link,'published_date':published_date,'title':title,'media':media})
if anchor_insert_status == 1:
print(f"First Anchor vertex INSERTED successfully")
# Create associated edges
country_edge_status = conn.upsertEdge("COUNTRY", country, "country_is", "ANCHOR_NEWS", id)
if country_edge_status == 1:
print(f"Country edge between {country} and anchor news {title} success")
topic_edge_status = conn.upsertEdge("TOPIC", topic, "topic_is", "ANCHOR_NEWS", id)
if topic_edge_status == 1:
print(f"Topic edge between {topic} and anchor news {title} success")
keyword1_edge_status = conn.upsertEdge("KEYWORD", keyword1, "keyword_belongs_to", "ANCHOR_NEWS", id)
if keyword1_edge_status == 1:
print(f"Keyword edge between {keyword1} and anchor news {title} success")
keyword2_edge_status = conn.upsertEdge("KEYWORD", keyword2, "keyword_belongs_to", "ANCHOR_NEWS", id)
if keyword2_edge_status == 1:
print(f"Keyword edge between {keyword2} and anchor news {title} success")
keyword3_edge_status = conn.upsertEdge("KEYWORD", keyword3, "keyword_belongs_to", "ANCHOR_NEWS", id)
if keyword3_edge_status == 1:
print(f"Keyword edge between {keyword3} and anchor news {title} success")
# Following news articles
else:
# Semantic search and cosine similarity between current news article and all existing anchor news articles
# This is where deduplication is done and a particular news article
# is classified as 'Anchor News' or 'Related News'
query_article_embedding = embedder.encode(summary)
anchor_embeddings = embedder.encode(anchor_vertices)
cosine_similarities = util.cos_sim(query_article_embedding,anchor_embeddings)[0]
print(cosine_similarities)
closest_to = -1
sim = 0
index = 0
for value in cosine_similarities:
if float(value) > THRESHOLD and float(value) > sim:
closest_to = index
sim = float(value)
index += 1
if closest_to == -1: # is a new 'anchor article'
anchor_vertices.append(summary)
anchor_to_children[id] = []
anchor_insert_status = conn.upsertVertex("ANCHOR_NEWS",id,{'rank':rank,'summary':summary,'link':link,'published_date':published_date,'title':title,'media':media})
if anchor_insert_status == 1:
print(f"Anchor vertex {title} INSERTED successfully")
country_edge_status = conn.upsertEdge("COUNTRY", country, "country_is", "ANCHOR_NEWS", id)
if country_edge_status == 1:
print(f"Country edge between {country} and anchor news {title} success")
topic_edge_status = conn.upsertEdge("TOPIC", topic, "topic_is", "ANCHOR_NEWS", id)
if topic_edge_status == 1:
print(f"Topic edge between {topic} and anchor news {title} success")
keyword1_edge_status = conn.upsertEdge("KEYWORD", keyword1, "keyword_belongs_to", "ANCHOR_NEWS", id)
if keyword1_edge_status == 1:
print(f"Keyword edge between {keyword1} and anchor news {title} success")
keyword2_edge_status = conn.upsertEdge("KEYWORD", keyword2, "keyword_belongs_to", "ANCHOR_NEWS", id)
if keyword2_edge_status == 1:
print(f"Keyword edge between {keyword2} and anchor news {title} success")
keyword3_edge_status = conn.upsertEdge("KEYWORD", keyword3, "keyword_belongs_to", "ANCHOR_NEWS", id)
if keyword3_edge_status == 1:
print(f"Keyword edge between {keyword3} and anchor news {title} success")
simscore_edge_status = conn.upsertEdge("ANCHOR_NEWS", "59671cd36eb3fd8c1a67853e55773599", "similarity_score", "ANCHOR_NEWS", id, {'score':id_to_semantic[id]})
if simscore_edge_status == 1:
print(f"Similarity score edge success")
else: # is a 'related article'
anchor_id = summary_to_id[anchor_vertices[closest_to]]
anchor_to_children[anchor_id].append(id)
related_insert_status = conn.upsertVertex("RELATED_NEWS",id,{'rank':rank,'summary':summary,'link':link,'published_date':published_date,'title':title,'media':media})
if related_insert_status == 1:
print(f"Related news vertex {title} INSERTED successfully")
country_edge_status = conn.upsertEdge("COUNTRY", country, "country_is", "RELATED_NEWS", id)
if country_edge_status == 1:
print(f"Country edge between {country} and related news {title} success")
topic_edge_status = conn.upsertEdge("TOPIC", topic, "topic_is", "RELATED_NEWS", id)
if topic_edge_status == 1:
print(f"Topic edge between {topic} and related news {title} success")
relation_edge_status = conn.upsertEdge("ANCHOR_NEWS", anchor_id, "is_related", "RELATED_NEWS", id)
if relation_edge_status == 1:
print(f"Relation edge between {anchor_id} and related news {id} success")
len(anchor_vertices) # number of anchor news or Distinct news
"""All data has been loaded successfully. View the GraphStudio to confirm, see connections.
## View of the graph from a **Country** perspective
# Visualization and Inference
"""
# Exploring graph
# -- Displaying vertex types and edge types in the graph
vertices = conn.getVertexTypes()
print(f"Vertices: {vertices}")
edges = conn.getEdgeTypes()
print(f"Edges: {edges}")
# Exploring individual vertex type and edge type
print(f"Results for Country vertex")
pprint(conn.getVertexType("COUNTRY"))
print()
print(f"Results for Topics edge")
pprint(conn.getEdgeType("topic_is"))
# Counting Graph Data
print("Vertex Counts")
for vertex in vertices:
print(f"There are {conn.getVertexCount(vertex)} {vertex} vertices in the graph")
print("--------------")
print("Edge Counts")
for edge in edges:
print(f"There are {conn.getEdgeCount(edge)} {edge} edges in the graph")
# Getting Individual Vertex
vertex_results = conn.getVerticesById("COUNTRY", "US") # Details about 'COUNTRY' vertex called 'US' (sentiment attribute values have also been rightly returned)
pprint(vertex_results)
# Count Edges Connected to a Particular Vertex
results = conn.getEdgeCountFrom("COUNTRY", "US")
pprint(results)
results = conn.getEdgeCountFrom("ANCHOR_NEWS", "59671cd36eb3fd8c1a67853e55773599") # one of the anchor news
pprint(results)
# Display all Edges Connected to a Particular Vertex
r1 = conn.getEdges("RELATED_NEWS", "87d8e333c13733a8105abc92d7d04c7b")
pprint(r1)
# Vertex set to dataframe
df3 = conn.vertexSetToDataFrame(vertex_results)
print(df3)
df3
# edge set to dataframe
df4 = conn.getEdgesDataframe("RELATED_NEWS", "87d8e333c13733a8105abc92d7d04c7b")
df4
# Pathfinding
shortest_path_results = conn.shortestPath([("COUNTRY", "US")], [("COUNTRY", "ES")])
pprint(shortest_path_results)
""""""
# Pathfinding
all_path_results = conn.allPaths([("COUNTRY", "US")], [("COUNTRY", "ES")],maxLength=6)
pprint(all_path_results)
"""Visual Output in TigerGraph Studio for the above All Path finding Query:
"""
# Retrieving all anchor vertices
query_result = conn.gsql('''
USE GRAPH Gemini
INTERPRET QUERY () SYNTAX V2{
anchors = SELECT a FROM ANCHOR_NEWS:a
ORDER BY a.rank ASC;
PRINT anchors[anchors.link,anchors.media,anchors.rank,anchors.summary,anchors.title];
}
''')
pprint(query_result)
query_result
"""### Analyzing Country sentiment for given news"""
# Running Interpreted queries
country_result = conn.gsql('''
USE GRAPH Gemini
INTERPRET QUERY () SYNTAX V2{
countries = SELECT c FROM COUNTRY:c
ORDER BY c.negative DESC;
PRINT countries[countries.negative];
}
''')
pprint(country_result)
country_result
# Create query
results = conn.gsql('''
USE GRAPH Gemini
CREATE QUERY country_negatives() FOR GRAPH Gemini SYNTAX V2{
countries = SELECT c FROM COUNTRY:c
ORDER BY c.negative DESC;
PRINT countries[countries.negative];
}
''')
pprint(results)
# installing queries
conn.gsql('''
INSTALL QUERY country_negatives
''')
# Run the installed query
results3 = conn.runInstalledQuery("country_negatives")
pprint(results3)
country_sentiment_df = pd.DataFrame(results3[0]['countries'])
country_sentiment_df
"""
**Conclusion** from above results: Many articles from US have a negative sentiment related to the search term/news.
### Similarly, you can run many more unique queries to mine valuable information from the data.
# Overall Summary
We now have separated initial huge number of news articles into 'anchor news' and 'related news' respectively. A powerful thing we can now do is combine the anchor news to give a summary of the whole topic/news at hand. This way, we are able to consolidate all viewpoints/perspectives from different news articles into one long-form passage and give it to the reader.
"""
# First, let us retrieve all the anchor news articles from our dataset
# -- we have to create a query, install and run it.
anchors = conn.gsql('''
USE GRAPH Gemini
CREATE QUERY get_anchors() FOR GRAPH Gemini SYNTAX V2{
anchors = SELECT a FROM ANCHOR_NEWS:a
ORDER BY a.rank ASC;
PRINT anchors[anchors.link,anchors.media,anchors.rank,anchors.summary,anchors.title];
}
''')
pprint(anchors)
# installing queries
conn.gsql('''
INSTALL QUERY get_anchors
''')
# run installed query
result_anchors = conn.runInstalledQuery("get_anchors")
pprint(result_anchors)
# Convert to dataframe
anchors_df = pd.DataFrame(result_anchors[0]['anchors'])
print(anchors_df)
anchors_df['attributes'][0]
"""Anchor news retrieval was successful"""
# Summarized result
from IPython.display import Image
for index, row in anchors_df.iterrows():
attributes = row['attributes']
summary = attributes['anchors.summary']
media = attributes['anchors.media']
link = attributes['anchors.link']
title = attributes['anchors.title']
print(f"{index+1}. {title}")
print(f"Link: {link} \n")
try:
display(Image(media,height=200,width=300))
except:
pass
print(f"\nSummary:\n{summary}")
print()
"""## Algorithms For More Insights
## Centrality Algorithm
-- Can be used to determine the order in which the anchor news articles (hyper nodes) are important. This will be the order in which the news articles are shown to the reader on the website.
"""
# Centrality Algorithm implementation
# -- we are implementing a specific type called "degree centrality"
centrality_query = conn.gsql('''
USE GRAPH Gemini
CREATE QUERY tg_degree_cent(SET<STRING> v_type, SET<STRING> e_type, SET<STRING> re_type, BOOL in_degree = TRUE, BOOL out_degree = TRUE,
INT top_k=100, BOOL print_accum = TRUE, STRING result_attr = "",STRING file_path = "") SYNTAX V1 {
TYPEDEF TUPLE<VERTEX Vertex_ID, FLOAT score> Vertex_Score;
HeapAccum<Vertex_Score>(top_k, score DESC) @@top_scores_heap;
SumAccum<INT> @sum_degree_score;
FILE f (file_path);
all = {v_type};
sll = SELECT s
FROM all:s
ACCUM IF in_degree THEN
FOREACH edge_type in re_type DO
s.@sum_degree_score+=s.outdegree(edge_type)
END
END,
IF out_degree THEN
FOREACH edge_type in e_type DO
s.@sum_degree_score+=s.outdegree(edge_type)
END
END;
IF file_path != "" THEN
f.println("Vertex_ID", "Degree");
END;
Start = SELECT s
FROM all:s
POST-ACCUM
IF result_attr != "" THEN
s.setAttr(result_attr, s.@sum_degree_score)
END,
IF print_accum THEN
@@top_scores_heap += Vertex_Score(s, s.@sum_degree_score)
END,
IF file_path != "" THEN
f.println(s, s.@sum_degree_score)
END;
IF print_accum THEN
PRINT @@top_scores_heap AS top_scores;
END;
}
''')
pprint(centrality_query)
# installing queries
conn.gsql('''
INSTALL QUERY tg_degree_cent
''')
# run installed query
# Use double quotes for datatype string while using gsql
centrality_result = conn.gsql('RUN QUERY tg_degree_cent(["ANCHOR_NEWS"], ["similarity_score","is_related","country_is","topic_is","keyword_belongs_to"],["similarity_score","is_related","country_is","topic_is","keyword_belongs_to"])')
pprint(centrality_result)
centrality_result2 = centrality_result.replace("\n","")
print(centrality_result2)
centrality_result3 = json.loads(centrality_result2)
centrality_result3
# Convert to dataframe
centrality_df = pd.DataFrame(centrality_result3['results'][0]['top_scores'])
centrality_df
"""Analysis successful. Above is the order of importance of the hyper-nodes in our graph.
## Similarity Algorithm
Earlier we had observed the take of US articles on our search term. Let us use **Similarity Algorithms** to determine which countries have an opinion similar to the US and which ones have a different one.
"""
# similarity algorithm
# -- we are implementing a specific kind called "Cosine Similarity of Neighborhoods (All Pairs, Batch)"
similarity_query = conn.gsql('''
USE GRAPH Gemini
CREATE QUERY tg_cosine_nbor_ap_batch(STRING vertex_type, STRING edge_type, STRING edge_attribute, INT top_k,
BOOL print_accum = true, STRING file_path, STRING similarity_edge, INT num_of_batches = 1) SYNTAX V1 {
TYPEDEF TUPLE<VERTEX ver, FLOAT val> res_tup; # storing results in the heap
MapAccum<VERTEX, FLOAT> @numerator_map, @@norm_map; # weight value from edge, normalized value
HeapAccum<res_tup>(top_k, val desc) @heap; # stores topK similarity results
FILE f (file_path);
start = {vertex_type};
subjects = SELECT t
FROM start:s -(edge_type:e)- :t
WHERE 4 > 0
ACCUM t.@numerator_map += (s -> 4),
@@norm_map += (s -> pow(4, 2));
FOREACH i IN RANGE[0, num_of_batches-1] DO
neighbours = SELECT t
FROM subjects:s -(edge_type:e)- :t
WHERE 4 > 0
AND getvid(t) % num_of_batches == i
ACCUM
FOREACH (k,v) IN s.@numerator_map DO
CASE WHEN getvid(k) != getvid(t) THEN # for testing purposes, using !=. Use > comparison operation normally
t.@numerator_map += (k -> v * 4)
END
END;
neighbours = SELECT t
FROM neighbours:t
POST-ACCUM
FOREACH (ver, w) IN t.@numerator_map DO
CASE WHEN ver != t THEN
FLOAT divisor = sqrt(@@norm_map.get(t) * @@norm_map.get(ver)),
CASE WHEN divisor == 0 THEN
CONTINUE
END,
FLOAT sim = w/divisor,
t.@heap += res_tup(ver, sim)
END
END,
t.@numerator_map.clear();
END;
start = SELECT t
FROM start:t
POST-ACCUM
FOREACH tup IN t.@heap DO
CASE WHEN tup.val > 0 THEN
IF file_path != "" THEN
f.println(t, tup.ver, tup.val)
END,
IF similarity_edge != "" THEN
INSERT INTO EDGE similarity_edge VALUES (t, tup.ver, tup.val)
END
END
END
ORDER BY getvid(t) ASC;
IF print_accum THEN
PRINT start[start.@heap];
END;
}
''')
pprint(similarity_query)
# installing queries
conn.gsql('''
INSTALL QUERY tg_cosine_nbor_ap_batch
''')
# run installed query
# Use double quotes for datatype string while using gsql
similarity_result = conn.gsql('RUN QUERY tg_cosine_nbor_ap_batch("COUNTRY", "country_is", "weight", 5, true, "", "", 1)')
pprint(similarity_result)
similarity_result2 = similarity_result.replace("\n","")
similarity_result3 = json.loads(similarity_result2)
similarity_df = pd.DataFrame(similarity_result3['results'][0])
similarity_df.head()
"""In a similar fashion, we can apply other types of algorithms like *Topological Link Prediction* and *classification* algorithms on our graph for gaining new types of knowledge.
#### Let us write some GSQL to retrieve mapping between anchor articles and their connected related_news articles. Will be useful for displaying information on the website in a structured way.
"""
# Data structures for storing required return data
anchor_vertices_list = []
anchor_to_related_news = {}
# Get anchors
result_anchors = conn.runInstalledQuery("get_anchors")
# Convert to dataframe
anchors_df = pd.DataFrame(result_anchors[0]['anchors'])
anchors_df
# Create and install our query
children = conn.gsql('''
USE GRAPH Gemini
CREATE QUERY get_children(VERTEX<ANCHOR_NEWS> news_id) FOR GRAPH Gemini SYNTAX V2{
Start = {news_id};
children = SELECT c FROM Start:a - () - RELATED_NEWS:c;
PRINT children[children.id,children.link,children.media,children.rank,children.summary,children.title];
}
''')
pprint(children)
# installing queries
conn.gsql('''
INSTALL QUERY get_children
''')
# run installed query
result_children = conn.runInstalledQuery("get_children",params={"news_id": "0e1632d42b5bc1a70f6692690bd2c783"})
pprint(result_children)
# Convert to dataframe
children_df = pd.DataFrame(result_children[0]['children'])
children_df
children_news_info = {}
# populate above created data structures
for index, row in anchors_df.iterrows():
anchor_id = row['v_id']
link = row['attributes']['anchors.link']
media = row['attributes']['anchors.media']
rank = row['attributes']['anchors.rank']
summary = row['attributes']['anchors.summary']
title = row['attributes']['anchors.title']
children_ids_list = []
children_results = conn.runInstalledQuery("get_children",params={"news_id": anchor_id})
temp_df = pd.DataFrame(children_results[0]['children'])
for index2,row2 in temp_df.iterrows():
children_ids_list.append(row2['v_id'])
c_id = row2['v_id']
c_link = row2['attributes']['children.link']
c_media = row2['attributes']['children.media']
c_rank = row2['attributes']['children.rank']
c_summary = row2['attributes']['children.summary']
c_title = row2['attributes']['children.title']
children_news_info[c_id] = {'link':c_link,'media':c_media,'rank':c_rank,'summary':c_summary,'title':c_title}
anchor_vertices_list.append({'id':anchor_id,'link':link,'media':media,'rank':rank,'summary':summary,'title':title,'children':children_ids_list})
len(anchor_vertices_list)
children_news_info
"""Intended result obtained. Save and download this data. We will use this in the website."""
# save data
import pickle
with open('anchor_vertices_list','wb') as file:
pickle.dump(anchor_vertices_list, file)
with open('children_news_info','wb') as file:
pickle.dump(children_news_info, file)
# load data
import pickle
with open('anchor_vertices_list','rb') as file:
anchor_vertices_list2 = pickle.load(file)
print(len(anchor_vertices_list2))