suchith720 / GraphWikipedia

This is code base for creating GraphWikiCategory and GraphWikiSeeAlso datasets.

GraphWikipedia Dataset.

This is a code base for creating GraphWikiCategory and GraphWikiSeeAlso dataset which was created from the wikipedia dumps, these datasets were primarily created for using Extreme classification with graphs.

Run all you code from the root directory of the repository, the steps to download the dataset are as follows:

1. Provide the following variables in the configurate file present at config/download_config.py

project = 'enwiki'
dump_date = "20220420"

#root directory of the repository
dataset_home = '/home/scai/phd/aiz218323/scratch/XML/wikipedia-data-science_2'

2. Run the download script with any of the following option:

• To get the dates of the wikipedia dumps python download.py --dates
• To download the wikipedia xml dumps python download.py --dumps
• To download the abstract dumps python download.py --abstract

For creating datasets from the dumps, we will need to process the wikitext present in the XML dumps for which we will be using the mwparserfromhell_2 library, which I have modified to accomodate few of our requirements. So add the path to this folder provided in the repository to $PYTHONPATH as follows :
PYTHONPATH='/path/to/GraphWikipedia/mwparserfromhell_2/src/:$PYTHONPATH

GraphWikiSeeAlso

This is article-to-article recommendation datasets, the classification task is given an article we need to predict the articles present in its Seealso section. Click here to download the dataset.

To create the dataset do the following:

1. Provide the following variables in the configurate file present at config/seealso_config.py

project = 'enwiki'
dump_date = "20220420"

#root directory of the repository
dataset_home = '/home/scai/phd/aiz218323/scratch/XML/wikipedia-data-science_2'

#saving the dumps using the download scripts
dump_dirname = 'raw_data'

#this is where datasets GraphWikiSeealso and GraphWikiCategory will be stored.
dataset_dir = f'{dataset_home}/datasets/'

#to the combined abstract information
abstract_dir = f'{dataset_dir}/Abstract/'

#to save the GraphWikiSeeAlso information
limit = None
category_dir = f'{dataset_dir}/WikiSeeAlso'
results_dir = f'{category_dir}/results'
combine_dir = f'{category_dir}/combined'
xc_dir = f'{category_dir}/XCData'

2. Run the following commands:

$ python src/generate_seealso_graph.py
$ python src/process_abstract_dumps.py
$ python src/create_graph-wiki-seealso.py

Dataset statistics

properties	no. of labels	no. of train	no. of test	points per label	labels per point
GraphWikiSeeAlso	333,411	668,400	293,023	4.215	2.10220

Graph statistics

properties	values
Number of nodes	6,162,086
Number of edges	165,199,765
Average in-degree	30.194
Average out-degree	26.809

Baselines

Here are some of the Extreme classification baselines:

Algorithm	P1	P3	P5	N1	N3	N5	PSP1	PSP3	PSP5	PSN1	PSN3	PSN5	MODELSIZE	TRNTIME	PREDTIME
Parabel	23.18	13.53	9.66	23.18	23.63	24.48	12.51	14.51	15.91	12.51	14.27	15.24	0.64	0.08	1.31
Bonsai	24.14	14.28	10.27	24.14	24.75	25.72	13.65	16.15	17.84	13.65	15.79	16.94	385.57	1466.05	22.47
PfastreXML	19.45	12.05	8.79	19.45	20.73	21.76	13.86	15.41	16.66	13.86	15.33	16.25	7.11	0.96	6.67
AnneXML	0.03	0.02	0.02	0.03	0.04	0.04	0.01	0.01	0.01	0.01	0.01	0.01	4449.60	1262.65	0.30
XT	22.14	13.19	9.55	22.14	22.88	23.89	12.01	14.28	15.95	12.01	13.95	15.09	2.1	12430.00	15.68

GraphWikiCategory

The classification task here is to tag a wikipedia article with the categories it belongs to. Wikipedia have special pages called category pages, when a wikilink to these pages is present in an article, the article is considered to belong to that particular category. Click here to download the dataset.

Here are the steps to create this dataset :

1. Provide the following variables in the configurate file present at config/category_config.py

project = 'enwiki'
dump_date = "20220420"

#root directory of the repository
dataset_home = '/home/scai/phd/aiz218323/scratch/XML/wikipedia-data-science_2'

#saving the dumps using the download scripts
dump_dirname = 'raw_data'

#this is where datasets GraphWikiSeealso and GraphWikiCategory will be stored.
dataset_dir = f'{dataset_home}/datasets/'

#to the combined abstract information
abstract_dir = f'{dataset_dir}/Abstract/'

#to save the GraphWikiCategory information
limit = None
category_dir = f'{dataset_dir}/WikiCategory'
results_dir = f'{category_dir}/results'
combine_dir = f'{category_dir}/combined'
xc_dir = f'{category_dir}/XCData'

2. Run the following commands:

$ python src/generate_category_graph.py
$ python src/process_abstract_dumps.py
$ python src/create_graph-wiki-category.py

Dataset statistics

properties	no. of labels	no. of train	no. of test	points per label	labels per point
GraphWikiCategory	1,080,148	4,245,310	1,830,068	26.387	4.698

Graph statistics

Article graph

properties	values
Number of nodes	6,471,181
Number of edges	206,162,248
Average in-degree	34.745
Average out-degree	31.858

Category graph

properties	values
Number of nodes	1,627,757
Number of edges	4,097,826
Average in-degree	5.5262
Average out-degree	2.5174

Baselines

Algorithm	P1	P3	P5	N1	N3	N5	PSP1	PSP3	PSP5	PSN1	PSN3	PSN5	MODELSIZE	TRNTIME	PREDTIME
Parabel	42.70	23.57	16.59	42.70	32.82	31.05	16.07	15.66	15.68	16.07	17.32	18.57	5.85	1.10	1.21
Bonsai	43.08	23.67	16.63	43.08	33.04	31.27	17.05	16.56	16.57	17.05	18.29	19.60	3492.86	29153.10	29.26
PfastreXML	36.24	19.62	13.90	36.24	27.83	26.54	18.52	15.22	14.80	18.52	17.66	18.59	48.73	10.16	12.22
AnneXML	42.19	22.54	15.83	42.19	31.86	30.12	15.28	14.25	14.41	15.28	16.16	17.44	27467.19	27704.24	0.12
XT	40.17	22.14	15.76	40.17	30.71	29.17	14.20	14.01	14.27	14.20	15.44	16.76	6.5	57169.00	8.77