Palantir and the Dynamic OntologyOrigin of the term ‘Ontology’An Ontology is a categorization of the worldWe frequently have created our own ontology whether we realize it or not…this will become clearer as we go
One of the earliest Ontological concepts came from Plato: Platonic realismPR is the idea that everything that exists in reality has a corresponding conceptual truth that the extant thing is aspiring to beThere is a universal notion of ‘apple-ness’ that a real apple aspires to beThis is one implementation of the idea that things in reality can have conceptual counterparts that are used to categorize themThis framework is mimicked in most Ontologies
One example of this ontological structure is the periodic tableThe squares in the table are not the actual element, they are just conceptualizations of the actual elementThese categorizations make it easier to interact with complex realities
Another well known Ontology is the Linnaean taxonomy or Latin naming system for animalsThis is a useful example of an Ontology because we can see how Ontologies are frequently ordered hierarchicallyA hierarchy is a nice way to order concepts.Again, note that the actual taxonomy of an animal is not the animal itself…there is an ethereal conceptualization and corporeal actualization Other Ontologies- Dewey Decimal system,
These same concepts can be (and are) applied to other domainsFor example, when organizations model their world, they care about how different objects (notably people) fit into a hierarchy
There are many ways to build an Ontology and we have already talked about a coupleIn general Ontologies have up to three basic information carriers: Objects, Properties, RelationshipsAlso referred to as ‘Actors’, ‘Attributes’ and ‘Links’ especially in SNACan also talk about Nodes and Edges to talk about Objects and LinksIn Palantir we allow for the use of all three information types.We will talk more about these information types in just a momentPause here to look at the Palantir Ontology using the Data Importer InterfaceData import is the point where outside information is mapped into the Ontology…one of many places in the product where we interact with the OntologyLook at Objects through table definitionLook at properties through mapping fieldsLook at links through the link type fields
Palantir is agnostic to an organization’s data modelThis is allowed by our Dynamic OntologyObjects are generally whatever is being modeledProperties are the qualities of those objectsRelationships are the connections between objectsDocuments are text basedEntities are generally person, place and thingEvents are things that occur in timeOnly the Ontology structure is hard-coded, but this is a very flexible structure that can accommodate almost any data model
Totally General to Totally Specific OntologyOne models nothing and one (attempts) to model everything – neither is an ideal way to model informationTotally GeneralAn almost empty ontology, information really is just an object, property or relationshipThe lack of semantics makes comparisons meaningless because apples and oranges truly are identicalTotally SpecificTry to model everything in the world in every domainIs highly cumbersome- makes the task of information mapping nearly impossible and frequently it is difficult to tell how a particular piece of information fits because frequently there will be more than one information type defined in the Ontology that could fit a piece of input datumWhat we want is an Ontology that can be as specific as it needs to be but has no ambiguity because it only includes the information relevant to a specific organization.Hard-coding this for every organization becomes a daunting task
Clearly a dynamic ontology allows an organization to model domain-relevant informationAnother huge advantage of a soft-coded ontology is that frequently similar concepts can be modeled in many different waysAn organization is not forced to model concepts in any particular wayConsider the modeling of a person’s occupation
Occupation can be modeled as an object definitionIn this model a person is a pilot, lawyer or doctor but not more than one of these thingsThis is useful if there are many object specific properties …in this case medical sub-specialty could be a property of a doctor
Occupation can be modeled as a propertyThe value of the occupation property can be populated with many different valuesIn this case a person can have multiple occupations assuming the ontology supports property multiplicity (Palantir does)Giving a doctor-specific property is more complicated in this configuration
Occupation can be modeled as a relationshipDifferent relationship types can create lots of semantic informationA person can have unlimited occupations assuming the link-types exist to support themLinks generally don’t have their own properties
What we have been talking about so far is a soft-coded ontologyThis isn’t the same thing as dynamic but it enables the dynamism of the Palantir Dynamic OntologyThe Palantir Ontology is dynamic because it can be changedIt can adjust as an organizations view of the world shifts
Unused objects, properties, relationships can be removedNew can be addedAll can be modifiedA long list of allowed modifications – these modifications dramatically affect the functionality of these different elementsBriefly show the PDOM interface to solidify the creation/deletion/modification concepts
How is this modeled in practice?
– Latin taxonomy of animals
Objects and Properties
– Periodic Table (has implicit relationships)
Objects and Relationships
– Properties can be modeled as relationships to ‘data’ objects
Objects and Properties and Relationships
– How information can be modeled in Palantir
The Ontology in Palantir
The Ontology permeates almost every
function in the Palantir Workspace.
Consequently, a well-formed Ontology is critical
to effective analysis.
Graph and Discovery: Type Folders
Search Interaction: AddSearch Around
Data Import: TableTypeBrowser Overview
Property Visualization:Types Panel
The Dynamic Ontology
Palantir uses a ‘Dynamic’ Ontology…An Ontological
structure…not an Ontology.
What does Dynamic mean?
– It means that Palantir does not use a hard-coded Ontology.
• In Palantir the only hard-coded notions are for Objects, Properties and
• Objects are further divided into Documents, Entities and Events
Why soft-code an Ontology?
A hard-coded Ontology is inherently limiting
– Forces an organization into one of two
Another Dynamic Advantage
Palantir’s Dynamic Ontology provides tremendous flexibility
– Can model many different concepts
– Can model the same concepts in many ways
Consider the question:
How do we model a person’s
A person’s job-function
could be classified using
an object definition.
A person’s job-function could
be classified by defining the
person’s occupation property.
It is also possible to define a
person’s job function through
their relationships with other
people and objects.
But is that really Dynamic?
Having a soft-coded Ontology is clearly valuable
– Relevant Semantics for any Domain
However, if a soft-coded Ontology can’t be changed, it is not much
better than a hard-coded Ontology
This is where the ‘Dynamic’ really comes in
–In Palantir, Ontologies can be
modified after deployment
What sort of Modifications?
Any objects, property or relationship that is not in use can be
Any new object, property or relationship can be added
The functionality of existing objects, properties and relationships
can be modified
−New/Modify Label Generators
−New/Modify property allowances
−New/Modify link allowances
Through Palantir’s Dynamic Ontology, any organization can build
the data model that is appropriate for their domain.
This Ontology is not hard-coded but rather soft-coded.
The Ontology can evolve along with the organization which
represents real dynamism.