Unlocking Knowledge Graphs: From Basics to Real‑World Applications

What is a Knowledge Graph

A Knowledge Graph is a semantic network that represents entities as nodes and their relationships as edges. By storing facts as triples (subject‑predicate‑object), it enables reasoning beyond keyword matching, e.g., answering questions like “Who are James Watt’s alumni?”.

Research Dimensions

Knowledge engineering – ontology construction, rule‑based inference, knowledge extraction and fusion.

Natural language processing – information extraction, entity disambiguation, semantic parsing.

Databases – RDF stores, data integration, graph‑based storage.

Machine learning – graph embedding, representation learning, graph neural networks.

Knowledge Engineering

The core components are a knowledge base and an inference engine.

Domain ontology construction : formal specification of shared concepts for a specific domain.

Knowledge extraction : acquiring facts from massive data using information‑extraction techniques.

Knowledge fusion : aligning and merging multiple graphs into a coherent whole.

Data Models

RDF : represents facts as triples (subject‑predicate‑object).

RDFS : extends RDF with a schema layer (classes, properties, domains, ranges) to support simple hierarchical inference.

OWL : further extends RDFS with class disjointness, property transitivity and automated reasoning.

Knowledge Extraction Pipeline

Typical steps are:

Entity recognition (named‑entity detection).

Entity disambiguation (linking mentions to canonical entities).

Relation extraction (identifying predicates between entities).

Rule‑based methods use dictionaries and pattern matching; machine‑learning methods train models such as Maximum Entropy or Conditional Random Fields on annotated corpora to predict entity boundaries and types.

Artificial‑Intelligence and Big‑Data Perspective

Knowledge graphs bridge symbolic AI (logic‑based reasoning) and connectionist AI (deep learning). Recent work integrates graph‑based representation learning (e.g., TransE) with neural models to achieve cognitive reasoning.

In the big‑data context, a knowledge graph is a relational‑analysis model that captures hidden value among the 5Vs (Volume, Velocity, Variety, Value, Veracity). Typical applications include:

Graph‑based machine learning (TransE, GCN).

Graph databases (RDF stores such as gStore, Virtuoso; property‑graph stores like Neo4j, JanusGraph).

Graph‑computing systems (Pregel, GraphLab).

Graph‑mining algorithms (PageRank, community detection, influence propagation).

Systems Developed by PKU (Beijing University)

The research group has released several open‑source tools:

gStore : an RDF graph database that uses subgraph matching. Supports up to 5 billion triples on a single machine. Version 0.9.1 (~140 k lines of C++). Repository: https://github.com/pkumod/gStore.

gBuilder : an end‑to‑end platform for knowledge‑graph construction, covering schema design, structured/unstructured data extraction, and multi‑model fusion.

gAnswer : a natural‑language QA system built on subgraph matching. Repository: https://github.com/pkumod/gAnswer.

Additional tools: gStore Workbench (visual management), gCloud (cloud service), gMaster (distributed deployment for billions of triples).

Representative Application Scenarios

FinTech : multi‑level equity analysis and risk assessment via graph queries.

Government big data : integration of civil‑registry records for social‑relationship retrieval.

Smart inspection : personnel integrity profiling and network analysis.

Healthcare : disease‑symptom‑drug graph built from drug manuals for intelligent medical QA.

AI assistants : voice‑assistant and chatbot QA systems.

Weather & traffic : real‑time ingestion of weather data and rule‑based matching for early warnings.

Public‑security : multi‑dimensional case exploration and hidden relationship mining across transportation, internet and immigration data.