News monitoring is a relationship problem
A news article can mention several companies, describe different events, and connect them to a wider economic or regulatory context. Analysts need to understand those relationships before deciding whether a story matters. Entity recognition and sentiment are useful ingredients, but their value depends on how they connect to the underlying information need.
For Sentometrics, Gilles contributed to named-entity recognition, sentiment analysis, and information extraction using both LLMs and traditional model-training methods. The engagement also included work on a hybrid knowledge graph for ESG and compliance, a technology roadmap, and funding proposals.
Use each method where it earns its place
The technical scope brings established NLP and newer language-model capabilities into the same problem space. Entity recognition provides a way to identify organisations and other relevant actors. Information extraction describes what the text says about them. Sentiment analysis adds another perspective on how an article presents a company or event.
The choice between model training and LLM-based extraction is consequently a task decision. Different methods offer different trade-offs in flexibility, consistency, and operational effort. A useful architecture makes those choices around the information being captured rather than treating a single model family as the answer to every requirement.
A graph makes connections explicit
The knowledge-graph work adds a connected representation to the text-analysis layer. Instead of leaving information in unrelated article outputs, a graph can express the relationships between entities and the information gathered about them. The engagement involved building and enhancing this knowledge layer alongside the language-processing capabilities it supports.
For an ESG or compliance workflow, that representation offers a useful organising principle: follow the entity, inspect the relevant relationships, and return to the original news context. The business purpose is clearer information access and interpretation. A graph alone does not establish that an allegation is true or that a company meets a regulatory requirement.
Technical depth, connected to a roadmap
The contribution combines NLP implementation with knowledge-graph engineering and strategic planning. Sentometrics's work on large-scale media monitoring places those capabilities in a clear business context: help stakeholders find and interpret information about companies. Roadmapping connects that purpose with the technical priorities needed to develop the product.
This case illustrates how AI consulting can bridge a domain question and its technical foundations. News monitoring becomes more useful when extraction tasks, entity relationships, and product priorities reinforce each other. The same approach applies to market intelligence, supplier monitoring, and document-heavy business processes where the connections matter as much as the individual records.
