Lab: RDF programming with RDFlib
- RDF graph programming with RDFlib
- from rdflib import Graph, Namespace, URIRef, BNode, Literal
- from rdflib.namespace import RDF, FOAF, XSD
- from rdflib.collection import Collection
- Graph: add(), remove(), triples(), serialize(), parse(), bind()
Continue with the graph you created in Exercise 1.
Task: Continue to extend your graph:
- Michael Cohen was Donald Trump's attorney.
- He pleaded guilty for lying to Congress.
- Michael Flynn was adviser to Donald Trump.
- He pleaded guilty for lying to the FBI.
- He negotiated a plea agreement.
If you want, you can try to use properties and types from standard vocabularies like FOAF (friend-of-a-friend) and DC (Dublin Core), but this is something we will look at in later exercises.
Task: According to this FRONTLINE article, Gates', Cohen's and Flynn's lying were different and are described in different detail.
- How can you represent "different instances of lying" as triples?
- How can you modify your knowledge graph to account for this?
Task: It is possible to solve the task above without blank (or anonymous nodes). But to do so, you need to create a URI for each "instance of lying". This is a situation where blank nodes may be more suitable. Change your graph so it represents instances of lying as blank nodes.
Task: Save (serialize) your graph to a Turtle file. Add a few triples to the Turtle file with more information about Donald Trump. For example, you can add that Donald Trump is married to Melania and has several children. You can also use blank nodes to represent two of Trump's addresses when he was president:
- The White House, 1600 Pennsylvania Ave., NW Washington, DC 20500, United States, phone: 1-202-456-1414
- Mar-a-Lago Club, 1100 S Ocean Blvd, Palm Beach, FL 33480, United States
Visualise the result if you want. Read (parse) the Turtle file back into a Python program, and check that the new triples are there.
If you have more time...
Task: Write a method (function) that starts with Donald Trump prints out a graph depth-first to show how the other graph nodes are connected to him. An excerpt of the output could be:
ex:Donald_Trump <== ex:campaignManager ex:Paul_Manafort ==> ex:convictedFor ex:BankAndTaxFraud ... <== ex:attorneyFor ex:Michael_Cohen ==> ex:pleadedGuilty ex:LyingToCongress
Here, the <== and ==> arrows are printed to indicate the reverse of a property. We do that with a print() statement in Python, not from inside rdflib.
Note: Because you must follow triples in both subject-to-predicate and predicate-to-subject direction, you must keep a list of already visited nodes, and never return to a previously visited one.
Note: If you want a neat solution, it may be best to combine two graph traversals: first traverse the model breadth-first to create a new tree-shaped model, and then traverse the tree-shaped model depth-first to print it out with indentation. (The point of the first breadth-first step is to find the shortest path to each node.)
Triples you can extend for the tasks (turtle format)
@prefix ex: <http://example.org/> . ex:Mueller_Investigation ex:involved ex:George_Papadopoulos, ex:Michael_Cohen, ex:Michael_Flynn, ex:Paul_Manafort, ex:Rick_Gates, ex:Roger_Stone ; ex:leadBy ex:Robert_Muller . ex:Paul_Manafort ex:businessManager ex:Rick_Gates ; ex:campaignChairman ex:Donald_Trump ; ex:chargedWith ex:ForeignLobbying, ex:MoneyLaundering, ex:TaxEvasion ; ex:convictedFor ex:BankFraud, ex:TaxFraud ; ex:negoiated ex:PleaBargain ; ex:pleadGuiltyTo ex:Conspiracy ; ex:sentencedTo ex:Prison . ex:Rick_Gates ex:chargedWith ex:ForeignLobbying, ex:MoneyLaundering, ex:TaxEvasion ; ex:pleadGuiltyTo ex:Conspiracy, ex:LyingToFBI .