project origin

context and challenges

Information management

spreadsheets, relational databases

Information extraction

turning text into structured data

Desire for automation

AI boom raised organizational awareness

outstanding needs

Lack of internal AI expertise

Difficulty in evaluating or negotiating services

Solutions not tailored to specific needs

Uncertain cost-benefit analysis

Lack of trust in LLMs

Hallucinations, opacity

interface between two canonical approaches

Top-down

symbolic rules

explicit specification of target regularities

High precision
Low recall
Explainable

Rigid
Requires few
examples

Bottom-up

deep learning, LLMs

implicitly captures regularities with numerous annotated examples

High recall
Enhanced handling
of context

Black box
Costly training
Limited ability to
explore data

hyperquest

best of both worlds

HyperQuest
reveals key patterns

Allows domain experts without AI expertise to select the most relevant cases

Suggests/co-constructs targeted information extraction rules with few examples

Expands assisted extraction through neuro-symbolic patterns

successive refinement of sub-corpora and increasingly relevant patterns

proposed solution

application ready to be used by non-specialists

Semantic Hypergraphs

generalize semantic graphs, great flexibility in defining regularities to extract

Ergonomic / No-code App

democratizes interactive extraction of relevant structured information

semantic hypergraph examples

Semantic Hypergraphs extend classical knowledge graphs by allowing edges to connect arbitrarily many nodes — and even other edges. This recursive structure captures the full expressiveness of natural language, including nested relations, qualifications, and epistemic context.