Context

Financial news contains entities, events, opinions, and statements whose usefulness depends on their relationships. A list of keywords cannot preserve all of that context. Gilles developed production NLP models and a hybrid knowledge graph for financial intelligence. The work covered entity recognition, sentiment, information extraction, ontology development, and processing. This is an engineering contribution to representing information; it does not establish investment performance, predictive validity, or a regulatory conclusion about the organizations mentioned in news.

Approach

Brass telephone patch panel with two connected leads

The contribution included designing and iterating the graph’s conceptual model, implementing extraction models, and developing the processing pipeline. Both trained models and LLMs were used for financial-news processing. A general graph approach makes the relationship between an entity, a statement, and its supporting material explicit. Event extraction from financial news provides a public research example of why participants and context matter. A useful representation distinguishes a statement about a company from a verified property of that company, particularly when reporting and opinion appear together.

Quality controls

An extracted relationship can be structurally valid while assigning the wrong participant or misreading a negation. General quality practice therefore separates entity errors, relation errors, and unsupported interpretations, retaining evidence for review. Model evaluation beyond valid JSON illustrates how reference data and explicit matching rules make those distinctions testable. These recommendations explain the engineering problem without asserting a particular unpublished benchmark, exhaustive validation suite, or measured accuracy for the commercial work.

Outcomes

Financial-news proof roll between steel guide rollers

The delivered outcome was production extraction and graph processing, supported by hands-on ontology and pipeline development. The public account establishes those capabilities and their information-processing purpose. It deliberately excludes client coverage figures, proprietary source inventories, operational volumes, and claims about downstream financial decisions.

  • Developed production entity, sentiment, and extraction models.
  • Built and iterated a production hybrid knowledge graph and its processing pipeline.