This month, we have discussed our most innovative project from a functional perspective, where we developed a solution that combines Knowledge Graphs with generative AI to transform the way we extract information and convert it into knowledge. In this post, we want to tell you about the process of knowledge construction, validation and the advantages of this technological approach.
We approach the process of knowledge building in three steps:
- In the first step, we digitise the document and perform a morphosyntactic analysis to extract relevant entities such as subjects, verbs, or predicates.
- Subsequently, in a second step, we perform semantic analysis to identify concepts and relationships such as axioms, determine hierarchies and restrictions, and incorporate them into the ontology.
- Finally, in the third step, we incorporate reasoning logic and business rules, obtaining a formal ontology capable of answering questions coherently.
Our validation process was inspired by game logic with a cooperative approach based on reinforcement learning, where the responses generated are validated by business experts and their feedback is incorporated into the system, achieving assisted learning that combines explicit and tacit knowledge. The KG+LLM paradigm offers qualitative and quantitative benefits over approaches based solely on embeddings. Specifically, the logical construction of knowledge allows us to define relationships between concepts more precisely and without hallucinations. Furthermore, the associative nature of ontologies allows us to extend knowledge and activate new domains through third-party ontologies.
In short, at WayOps we are convinced that thanks to this paradigm we are entering the era of reliable knowledge, where we can put an end to the hallucinations of Generative AI and extend its use cases to all areas of business. If your company is facing hallucinations that limit the adoption of Generative AI, please contact us so we can solve it together.
No hay comentarios