Mining neighbors, paths, and path patterns from a knowledge graph and a set of seed nodes

A detailed description of the motivation and the algorithms of kgpm is available in the related article.

When citing kgpm, please use the following reference:

Pierre Monnin, Emmanuel Bresso, Miguel Couceiro, Malika Smaïl-Tabbone, Amedeo Napoli, and Adrien Coulet. "Tackling scalability issues in mining path patterns from knowledge graphs: a preliminary study". In: 1st international conference "Algebras, graphs and ordered sets" (ALGOS 2020). Ed. by Miguel Couceiro, Pierre Monnin, and Amedeo Napoli. Nancy, France, Aug. 2020. url: https://arxiv.org/pdf/2007.08821.pdf.

@inproceedings{Monnin2020kgpm,
	author = {Monnin, Pierre and Bresso, Emmanuel and Couceiro, Miguel and Sma{\"i}l-Tabbone, Malika and Napoli, Amedeo and Coulet, Adrien},
	title = {{Tackling scalability issues in mining path patterns from knowledge graphs: a preliminary study}},
	editor    = {Miguel Couceiro and Pierre Monnin and Amedeo Napoli},
	booktitle = {{1st international conference ``Algebras, graphs and ordered sets'' (ALGOS 2020)}},
	address = {Nancy, France},
	year = {2020},
	month = Aug,
	url = {https://arxiv.org/pdf/2007.08821.pdf},
}

Python script to query a knowledge graph and perform its canonicalization. The script outputs:

• Files representing the canonical knowledge graph (in rdf_to_canonical_index, canonical_to_rdf_index, canonical_graph_adjacency, canonical_graph_inv_adjacency, rdf_nodes_cache_manager.csv, predicates_cache_manager.csv)
• Statistics about the knowledge graph before and after canonicalization (in graphs_statistics.md)

Parameters:

• --configuration: path of the JSON configuration file
• --max-rows: max number of rows the SPARQL endpoint can return
• --output: base directory for output files
• --self-signed-ssl: enable self signed SSL certificates
• --debug: print debug statements

Python script to mine neighbors, paths, and path patterns from a canonical knowledge graph and a set of seed nodes.

Parameters:

• --configuration: path of the JSON configuration file
• --graph: base directory for the input graph files
• --dataset-csv: CSV file with the seed nodes URIs (column 0) and class labels (column 1)
• --dataset-name: name of the data set
• --output: base directory for output files (statistics, scipy matrice of nodes x features, column name file, and a numpy vector of class labels)
• -d: maximum degree to allow expansion (disabled with d = -1)
• --lmin: minimum support for features
• --lmax: maximum support for features
• --kmin: minimum k to test (i.e., number of traversed edges, size of paths and path patterns)
• --kmax: maximum k to test
• --tmin: minimum t to test (i.e., level for generalization in class hierarchies); t = -1 disables type generalization, t = 0 only allows to generalize with owl:Thing
• --tmax: maximum t to test
• --undirected: whether only out arcs (false) or all arcs (true) are traversed
• --meaningful: biomedical additional filtering strategies:
  • p: only select features containing a pathway
  • g: only select features containing a gene or a GO class
  • m: only select features containing a MeSH class
  • pg: disjunction of p and g
  • pgm: disjunction of p, g, and m
  • all: test all previous filters (thus, 5 outputs)
  • no_check: disable the additional filtering
• --debug: print debug statements

Python script to compute the statistics about the subgraph accessible from a set of seed nodes in a canonical knowledge graph. It outputs a markdown file containing the number of neighbors and types reachable from the seed nodes.

Parameters:

• --configuration: path of the JSON configuration file
• --graph: base directory for the input graph files
• --dataset-csv: CSV file with the seed nodes URIs (column 0) and class labels (column 1)
• --dataset-name: name of the data set
• --output: base directory for output files (Markdown files)
• -d: maximum degree to allow expansion (disabled with d = -1)
• --undirected: whether only out arcs (false) or all arcs (true) are traversed
• --detailed: enable detailed statistics, i.e., number of neighbors and types accessible w.r.t. k and t until full neighborhood is reached. By default, only the max numbers of reachable neighbors and types in the full neighborhood are output (k and t are not given).

An example of a JSON configuration file is given in configuration.json.example. Keys are:

• server-address: address of the SPARQL endpoint to query
• url-json-conf-attribute: URL attribute to use to get JSON results
• url-json-conf-value: value of the url-json-conf-attribute to get JSON results
• url-default-graph-attribute: URL attribute to use to define the default graph
• url-default-graph-value: value of url-default-graph-attribute to define the default graph
• url-query-attribute: URL attribute to use to define the query
• timeout: timeout value for HTTP requests
• username: username to use if HTTP authentication is required (empty otherwise)
• password: password to use if HTTP authentication is required (empty otherwise)
• path_predicates_blacklist: blacklist of URIs or prefixes of predicates not to traverse
• types_blacklist: blacklist of URIs or prefixes of types not to use in path generalization
• types_expansion_blacklist: blacklist of URIs or prefixes of types whose instances cannot be traversed

• tqdm
• numpy
• bitarray
• scipy