A knowledge graph contains (Einstein, bornIn, Ulm) and (Ulm, locatedIn, Germany). Without any inference rules configured, what does a SPARQL query for 'Where was Einstein born?' return?
AGermany — the graph automatically applies transitivity to infer the broader location
BUlm — only explicitly stored triples are returned; inference requires explicitly configured rules or ontologies
CBoth Ulm and Germany — knowledge graphs return all logically derivable answers by default
DNothing — the query requires natural language processing unavailable in SPARQL
A knowledge graph is not an inference engine by default — it is a store of explicit triples. Returning 'Germany' requires an explicitly applied transitivity rule (e.g., via OWL reasoning or a SPARQL CONSTRUCT query). This is the key distinction between what is *stored* and what can be *inferred*. Students often assume KGs automatically reason over all implied facts; in practice, inference requires deliberate configuration of ontology rules.
Question 2 Multiple Choice
To find all colleagues of a person who have won a Nobel Prize, a knowledge graph query would:
APerform a two-table JOIN on a 'colleagues' table and an 'awards' table
BTraverse 'colleague' edges from the person node, then check each neighbor for 'wonAward' edges pointing to Nobel Prize nodes
CSearch all triples where the person appears as a subject
DQuery a single triple (person, nobelPrize, ?) directly
The power of the graph representation is multi-hop traversal: follow one type of edge, then follow another. This is natural in a graph (two hops along typed edges) but requires explicit JOINs in a relational model. Option A describes the relational approach; in a knowledge graph, the structure itself enables this pattern without schema-defined join tables. This is why graph queries can express relationship chains that would require multiple JOINs in SQL.
Question 3 True / False
Knowledge graph embeddings represent entities and relations as continuous vectors, enabling prediction of relationships that were never explicitly stored as triples.
TTrue
FFalse
Answer: True
Embedding methods (TransE, DistMult, ComplEx, etc.) learn vector representations such that the geometric relationship between entity and relation vectors encodes semantic relationships. A trained model can score candidate triples and predict likely missing links — for example, inferring that two drugs probably share a molecular target even if that fact isn't in the graph. This is how knowledge graphs bridge symbolic reasoning with statistical machine learning.
Question 4 True / False
A knowledge graph and a relational database are equivalent in what they can represent: both use tables of facts and support the same query operations.
TTrue
FFalse
Answer: False
While both can technically encode the same facts, knowledge graphs differ in key ways. The triple store's flexible schema allows adding new relation types without altering a table schema. More importantly, graph queries natively express arbitrary-depth path traversal (multi-hop relationships) without predefined joins, and ontologies (RDF Schema, OWL) enable symbolic inference over class hierarchies and property constraints. These capabilities make knowledge graphs especially suited to heterogeneous, evolving knowledge where relationships are themselves first-class objects.
Question 5 Short Answer
What role does entity resolution play in a knowledge graph, and why is it necessary when building large-scale graphs from multiple sources?
Think about your answer, then reveal below.
Model answer: Entity resolution (also called entity linking or deduplication) is the process of identifying when different names, identifiers, or descriptions in different data sources refer to the same real-world entity. For example, 'Albert Einstein,' 'A. Einstein,' and 'Einstein, Albert' in three different datasets must be recognized as the same node. Without entity resolution, the graph contains duplicate nodes for the same entity, breaking traversal and inference — a query about Einstein's publications would miss results from sources that used a different string. Large KGs like Wikidata solve this by assigning canonical entity IDs and maintaining aliases.
This tests whether students understand the practical engineering challenges of building a knowledge graph, not just its abstract structure. The key insight is that the power of a KG depends on the graph being connected correctly — duplicate nodes for the same entity sever the chains that enable multi-hop queries and inference. Entity resolution is what transforms a collection of independent triples into a coherent semantic network.