Semantic Web Provenance

Curt Tilmes

CMSC 691S

Provenance

• Provenance, from the French provenir, "to come from", means the origin, or the source, of something, or the history of the ownership or location of an object. [wikipedia://Provenance]
• A superset of Pedigree(lineage, ancestry), including a comprehensive set of information about where something came from and how it got from its creation to its current location or state.
• A very important concept in many fields: Fine Arts, Supply Chains, Food Supply, Production/Manufacturing Quality Analysis, Legal Chain of Evidence, Pharmaceutical Development, Science, etc.

Reputation, Validation, Certification

• Reputation is the opinion (more technically, a social evaluation) of the public toward a person, a group of people, or an organization. It is an important factor in many fields, such as education, business, online communities or social status. [wikipedia://Reputation]
• Validation is the process of checking if something satisfies a certain criterion.[wikipedia://Validation]
• Certification refers to the confirmation of certain characteristics of an object, person, or organization. This confirmation is often, but not always, provided by some form of external review, education, or assessment. [wikipedia://Certification]

Proof and Trust

• These are all different methods of establishing Trust.
• If you want to trust the outcome, you have to trust the inputs, and you have to trust the process.
• Proof is the justification and documentation of that process.
• The amount of proof needed is related to the degree of trust required by various classes of users. This is driven by the planned use of the information.

Proof and Trust

In the W3C stack, Proof builds on the other technologies and leads to Trust.

• A human can review a natural language proof, but a computer agent needs a semantic description of the proof.

Some provenance mechanisms

• RDF reification provides a mechanism for storing information about triples ("meta-information").
• N-Quads can be used to store the provenance of each triple by extending each triple with its source URI: subject-predicate-object-context.
• Named Graphs can use the source URI of the sets of triples as the name of the sub-graph they represent.
• Proof Markup Language (PML) is an OWL ontology that can represent provenance and justifications of reasoning results.

Proof Markup Language

PML (ver 2) is split into three ontology modules:

• The provenance ontology (PML-P) focuses on primitives for referencing and annotating entities.
• The justification ontology (PML-J) focuses on primitives for encoding justifications for how a conclusion was derived.
• The trust relation ontology (PML-T) focuses on primitives for explaining belief assertions associated with information and trust assertions associated with sources.

Example Scenario

John consults an online tour guide agent Agent X about the specialty of a local restaurant called Tonys.

John: What type of food is Tonys' specialty?
X: Tonys' specialty is Shellfish.

or, formally, in Knowledge Interchange Format (KIF):

Question:  (type TonysSpecialty ?x) 
Answer:    (type TonysSpecialty ShellFish)

Where did this answer come from? How much should I trust it?

Example Encoding in PML-P

Encode the question:

  <Question rdf:about="http://...">
    <pmlp:hasContent>
      <pmlp:Information>
        <pmlp:hasRawString rdf:datatype="http://www.w3.org/2001/XMLSchema#string">
          What type of food is Tony's specialty?
        </pmlp:hasRawString>
      </pmlp:Information>
    </pmlp:hasContent>
  </Question>

Encode information:

  <pmlp:SourceUsage> 
    <pmlp:hasSource> 
      <pmlp:DocumentFragmentByOffset> 
        <pmlp:hasDocument rdf:resource="http://..."/> 
        <pmlp:hasFromOffset rdf:datatype="http://www.w3.org/2001/XMLSchema#int">
          58
        </pmlp:hasFromOffset> 
        <pmlp:hasToOffset rdf:datatype="http://www.w3.org/2001/XMLSchema#int">
          91
        </pmlp:hasToOffset> 
      </pmlp:DocumentFragmentByOffset> 
    </pmlp:hasSource> 
    <pmlp:hasUsageDateTime rdf:datatype="http://www.w3.org/2001/XMLSchema#dateTime">
      2005-10-17T10:30:00Z
    </pmlp:hasUsageDateTime> 
  </pmlp:SourceUsage> 

Justifications

Four types:

• Type I: unproved conclusion or goal

    '(type TonysSpecialty SHELLFISH)' is either unproved or a goal
  

• Type II: the conclusion is an assumption

    '(type TonysSpecialty SHELLFISH)' has been directly assumed by the JTP
    inference engine
  

• Type III: the conclusion is a direct assertion

    '(type TonysSpecialty SHELLFISH)' has been directly asserted in
    "Stanford's Tonys Specialty Example" as a span of text between byte
    offset 58 and byte offset 91 as of 10:30 on 2005-10-17
  

Justifications

• Type IV: the conclusion is derived from a list of antecedents by applying a certain computation (e.g. inference rule)

  '(subClassOf CRAB SHELLFISH)' has being directly asserted in
  "Stanford's Tony's Specialty Ontology" as a span of text between byte
  offset 56 and byte offset 82 as of 10:30 on 2005-10-17
  
  and
  
  '(or (not(subClassOf CRAB ?x)) (type TonysSpecialty ?x))' has being
  directly asserted in "Deborah" as of 10:30 on 2005-10-17
  
  and 
  
  '(type TonysSpecialty SHELLFISH)' is derived from the application of
  General Modus Ponens rule on the two premises linked by
  http://.../answer_2/ns1.owl#ns1 and http://.../answer_2/ns2.owl#ns2,
  and we also record that the inference step binds the variable ?x to
  SHELLFISH.
  
  

Trust

• PML-T can assert simple trust statements like "Agent A believes information #B" and "Agent C trusts Agent D." It can also be extended to handle more complex trust models, such as propagating trust information or calculating trust.
• Wiki Trust Demo

References

• N-Quads: Extending N-Triples with Context
• Named Graphs as a Mechanism for Reasoning about Provenance
• A proof markup language for semantic web services
• PML 2: A Modular Explanation Interlingua
• Explaining Answers from the Semantic Web.
• Proof Markup Language (PML) Primer