Package RDFClosure

source code

This module is brute force implementation of the 'finite' version of RDFS semantics and of OWL 2 RL on the top of RDFLib (with some caveats, see below). Some extensions to these are also implemented. Brute force means that, in all cases, simple forward chaining rules are used to extend (recursively) the incoming graph with all triples that the rule sets permit (ie, the "deductive closure" of the graph is computed). There is an extra options whether the axiomatic triples are added to the graph (prior to the forward chaining step). These, typically set the domain and range for properties or define some core classes. In the case of RDFS, the implementation uses a 'finite' version of the axiomatic triples only (as proposed, for example, by Herman ter Horst). This means that it adds only those rdf:_i type predicates that do appear in the original graph, thereby keeping this step finite. For OWL 2 RL, OWL 2 does not define axiomatic triples formally; but they can be deduced from the OWL 2 RDF Based Semantics document and are listed in Appendix 6 (though informally). Note, however, that this implementation adds only those triples that refer to OWL terms that are meaningful for the OWL 2 RL case.

Package Entry Points

The main entry point to the package is via the DeductiveClosure class. This class should be initialized to control the parameters of the deductive closure; the forward chaining is done via the expand method. The simplest way to use the package from an RDFLib application is as follows:

       graph = Graph()                                 # creation of an RDFLib graph
       ...
       ...                                             # normal RDFLib application, eg, parsing RDF data
       ...
       DeductiveClosure(OWLRL_Semantics).expand(graph) # calculate an OWL 2 RL deductive closure of graph
                                                       # without axiomatic triples

The first argument of the DeductiveClosure initialization can be replaced by other classes, providing different types of deductive closure; other arguments are also possible. For example:

DeductiveClosure(OWLRL_Extension, rdfs_closure = True, axiomatic_triples = True, datatype_axioms = True).expand(graph)

will calculate the deductive closure including RDFS and some extensions to OWL 2 RL, and with all possible axiomatic triples added to the graph (this is about the maximum the package can do…)

The same instance of DeductiveClosure can be used for several graph expansions. In other words, the expand function does not change any state.

For convenience, a second entry point to the package is provided in the form of a function called convert_graph, that expects a directory with various options, including a file name. The function parses the file, creates the expanded graph, and serializes the result into RDF/XML or Turtle. This function is particularly useful as an entry point for a CGI call (where the HTML form parameters are in a directory) and is easy to use with a command line interface. The package distribution contains an example for both.

There are major closure type (ie, semantic closure possibilities); these can be controlled through the appropriate parameters of the DeductiveClosure class:

• using the RDFS_Semantics class, implementing the RDFS semantics
• using the OWLRL_Semantics class, implementing the OWL 2 RL
• using RDFS_OWLRL_Semantics class, implementing a combined semantics of RDFS semantics and OWL 2 RL

In all three cases there are other dimensions that can control the exact closure being generated:

• for convenience, the so called axiomatic triples (see, eg, the axiomatic triples in RDFS) are, by default, not added to the graph closure to reduce the number of generated triples. These can be controlled through a separate initialization argument
• similarly, the axiomatic triples for D-entailment are separated

Some Technical/implementation aspects

The core processing is done in the in the Core class, which is subclassed by the RDFS and the OWL 2 RL classes (these two are then, on their turn, subclassed by the RDFS + OWL 2 RL Semantics) class). The core implements the core functionality of cycling through the rules, whereas the rules themselves are defined and implemented in the subclasses. There are also methods that are executed only once either at the beginning or at the end of the full processing cycle. Adding axiomatic triples is handled separately, which allows a finer user control over these features.

Literals must be handled separately. Indeed, the functionality relies on 'extended' RDF graphs, that allows literals to be in a subject position, too. Because RDFLib does not allow that, processing begins by exchanging all literals in the graph for bnodes (identical literals get the same associated bnode). Processing occurs on these bnodes; at the end of the process all these bnodes are replaced by their corresponding literals if possible (if the bnode occurs in a subject position, that triple is removed from the resulting graph). Details of this processing is handled in the separate Literals Proxies class.

The OWL specification includes references to datatypes that are not in the core RDFS specification, consequently not directly implemented by RDFLib. These are added in a separate module of the package.

Problems with Literals with datatypes

The current distribution of RDFLib is fairly poor in handling datatypes, particularly in checking whether a lexical form of a literal is "proper" as for its declared datatype. A typical example is:

 "-1234"^^xsd:nonNegativeInteger

which should not be accepted as valid literal. Because the requirements of OWL 2 RL are much stricter in this respect, an alternative set of datatype handling (essentially, conversions) had to be implemented (see the XsdDatatypes module).

The DeductiveClosure class has an additional instance variable whether the default RDFLib conversion routines should be exchanged against the new ones. If this flag is set to True and instance creation (this is the default), then the conversion routines are set back to the originals once the expansion is complete, thereby avoiding to influence older application that may not work properly with the new set of conversion routines.

If the user wants to use these alternative lexical conversions everywhere in the application, then the use_improved_datatypes_conversions method can be invoked. That method changes the conversion routines and, from that point on, all usage of DeductiveClosure instances will use the improved conversion methods without resetting them. Ie, the code structure can be something like:

 DeductiveClosure().use_improved_datatypes_conversions()
 ... RDFLib application
 DeductiveClosure().expand(graph)
 ...

The default situation can be set back using the use_rdflib_datatypes_conversions call.

It is, however, not required to use these methods at all. Ie, the user can use:

 DeductiveClosure(improved_datatypes=False).expand(graph)

which will result in a proper graph expansion except for the datatype specific comparisons which will be incomplete.

Problems with Literals with datatypes

The current distribution of RDFLib is fairly poor in handling datatypes, particularly in checking whether a lexical form of a literal is "proper" as for its declared datatype. A typical example is:

 "-1234"^^xsd:nonNegativeInteger

which should not be accepted as valid literal. Because the requirements of OWL 2 RL are much stricter in this respect, an alternative set of datatype handling (essentially, conversions) had to be implemented (see the XsdDatatypes module).

The DeductiveClosure class has an additional instance variable whether the default RDFLib conversion routines should be exchanged against the new ones. If this flag is set to True and instance creation (this is the default), then the conversion routines are set back to the originals once the expansion is complete, thereby avoiding to influence older application that may not work properly with the new set of conversion routines.

If the user wants to use these alternative lexical conversions everywhere in the application, then the use_improved_datatypes_conversions method can be invoked. That method changes the conversion routines and, from that point on, all usage of DeductiveClosure instances will use the improved conversion methods without resetting them. Ie, the code structure can be something like:

 DeductiveClosure().use_improved_datatypes_conversions()
 ... RDFLib application
 DeductiveClosure().expand(graph)
 ...

The default situation can be set back using the use_rdflib_datatypes_conversions call.

It is, however, not required to use these methods at all. Ie, the user can use:

 DeductiveClosure(improved_datatypes=False).expand(graph)

which will result in a proper graph expansion except for the datatype specific comparisons which will be incomplete.

Imports: StringIO, IntType, TypeType, BooleanType, CodeType, UnboundMethodType, StringType, BuiltinMethodType, FloatType, DictionaryType, NotImplementedType, BuiltinFunctionType, DictProxyType, GeneratorType, InstanceType, ObjectType, DictType, GetSetDescriptorType, FileType, EllipsisType, StringTypes, ListType, MethodType, TupleType, ModuleType, FrameType, LongType, BufferType, TracebackType, ClassType, MemberDescriptorType, UnicodeType, SliceType, ComplexType, LambdaType, FunctionType, XRangeType, NoneType, rdflib, rdflibLiteral, Graph, DatatypeHandling, Closure, OWLRL_Extension, OWLRL_Extension_Trimming, OWLRL_Semantics, RDFS_Semantics, RDFS_OWLRL_Semantics, imports, JsonLDParser, JsonLDSerializer, register, Serializer, Parser, AxiomaticTriples, CombinedClosure, Literals, OWLRL, OWLRLExtras, RDFSClosure, RestrictedDatatype, XsdDatatypes