Knowledge Graph

Go beyond finding similar passages

Search can locate similar text, but similarity alone does not define whether one skill is a prerequisite for another or which equipment a process depends on. A graph makes selected relationships explicit. Research on GraphRAG also explores graph-based summaries for questions spanning a document collection; results depend on the graph and retrieval design.

The approach

A knowledge graph represents entities or concepts as nodes and relationships as edges. Technica AI uses business context to define which relationships matter, then combines the graph with other retrieval methods where appropriate.

1. Define the vocabulary: Agree concept types and relationships for the target business area.
2. Build candidate links: Extract or assemble relevant information from configured sources.
3. Validate the structure: Consolidate terminology and check important relationships with domain expertise.
4. Use the graph: Follow relevant links alongside text retrieval and show available evidence behind recommendations.

Business value

• Explore connected information

  Find dependencies and related concepts beyond the wording of a query.

• Support complex questions

  Gather context across connected entities, processes and documents.

• Explain recommendations

  Show relevant relationship paths so users can assess why a result was proposed.

Application considerations

• Graph quality depends on source information and how relationships are defined. Begin with a focused business area and agree maintenance responsibilities.
• In AIKNOW Matching, related skills indicate possible transferability to verify, not proven readiness. Profiles update automatically, but the ontology does not update automatically; people make final selection decisions.

Related products

• AIKNOW Matching

  Uses a configured skill ontology and recorded experience to recommend talent for projects and teams.

• AIKNOW

  Combines graph and other retrieval methods to connect enterprise knowledge.

• AISECI

  Provides a knowledge-creation framework whose target architecture includes combining dispersed knowledge.

References

• Hogan, A. et al. (2021). Knowledge Graphs. ACM Computing Surveys, 54(4). arXiv:2003.02320. https://arxiv.org/abs/2003.02320
• Edge, D. et al. (2024). From Local to Global: A Graph RAG Approach to Query-Focused Summarization. arXiv:2404.16130. https://arxiv.org/abs/2404.16130
• W3C (2009). SKOS Simple Knowledge Organization System Reference. W3C Recommendation. https://www.w3.org/TR/skos-reference/