chapter eight
8 NED with open LLMs and domain ontologies
This chapter covers
- Understanding the limitations of traditional NED tools
- Combining general-purpose LLMs and domain ontologies for NED
- Performing multistep disambiguation with shortest-paths detection, path-to-text translation, and textual paths summarization
Chapter 7 is focused on named entity disambiguation (NED), highlighting the role of scispaCy, a specialized natural language processing (NLP) tool built on the spaCy framework. This tool is designed to process documents and publications by providing pretrained models in the biomedical domain.
8.1 Understanding limitations of traditional NED systems
scispaCy incorporates vocabularies and ontologies, such as the Unified Medical Language System (UMLS), that provide canonical entities useful for disambiguating mentions in the text. However, this approach has some limitations:
- It is designed for a particular application domain: the biomedical field.
- It presents challenges in expanding and updating the reference knowledge base to incorporate new entities and terms.
- It fails to fully use the extensive information available in the knowledge base.
- It doesn’t use the existing relationships and paths between entities for the disambiguation task.
To understand the effect of this last point, let’s recap the example we discussed at the beginning of chapter 7, which used this quote from the European Centre for Disease Prevention and Control (ECDC):