Taxonomy, Thesaurus, or Ontology? – Clearing the Confusion, Part Two
By Marjorie Hlava, Chief Science Officer
Structured Vocabularies
In this part of our explanation of taxonomies, thesauri, and ontologies, we will refer to them collectively as “structured vocabularies.” They are also referred to as Knowledge Organization Systems (KOS). For a field which tries to contain and streamline language we certainly have a lot of variations in the naming!
Access Innovations builds structured vocabularies for our clients based on their information assets, industry, and needs. The structured vocabulary can be built from scratch or constructed using and customizing one or more of our 50-plus existing Knowledge Domains. Each Knowledge Domain is built around a specific topic or content area and can be formatted as a taxonomy or more fully formatted as a thesaurus with full records for every term.
Indexing Data
Knowledge Domains enhance knowledge management systems by enabling tagging (also known as data labeling or indexing), improving search accuracy, supporting web navigation, facilitating semantic fingerprinting, and more.
A thesaurus promotes search through tagging, which is done at the most specific term level. The content can be tagged as a whole piece/article, a chapter in a book or down to the paragraph and figure level. Searchers find documents or assets based on their tags. However, when tagging does not provide the full or right answer, we move to a full-text search using synonyms. Access uses Lucene/Solr or Endeca in search because they both support synonyms and full text search.
From there, Access Innovations can further build out the thesaurus into an ontology. We offer a full OWL export which is the W3C Web Ontology Language system. We also offer OWL Lite and OWL DL. In addition, we provide four different SKOS (Simple Knowledge Organization Systems) formats to build frameworks for association. SKOS does not have all the fields available that the thesaurus or ontology has. Therefore, we can export a knowledge asset base to any ontology system with all its terms, attributes, classes as well as relationships.
Ontologies
Ontologies have become more important with the use of artificial intelligence (AI). Access Innovations supports ontologies through our Data Harmony software suite, which leverages AI by improving the LLMs (large language models) it utilizes to present concepts with clarity, transparency, and reliability. Accuracy is a huge concern when deploying AI, and some of our clients have reported we have helped them improve data labeling accuracy to as high as 99.98%.
The tagging can be, as noted earlier, at the article or item level. It can also be “chunked” to the sentence, partial sentence or paragraph level as needed for ingestion. For large source text, it is preferable to tag at a more discrete level.
Accuracy
Accuracy through both factual lining and consistency in term usage is needed to minimize AI hallucination, especially in domain-specific topical areas. LLM’s also need ontologies to provide the semantic mapping of concepts and entities (people, places and things). The relationships between the term provide a structure for the content so the LLM can reason beyond the simple chunks and text patterns.
Unifying the data into a rich semantic topical layer provided by the tagging supports data governance and a reliable content base. The system is no longer just a black box of chunks and tokens but actually shows why an answer was given. The traceability fosters trust and explainable AI to the users.
Better automated workflows come from the ontologies guiding the LLM in fetching the relevant information from the knowledge bases in Retrieval-Augmented Generation (RAG) systems. The retrieval of information is more relevant.
Finally with tagged discrete pieces of content, chunks, the data becomes reusable and re-combinable for the creation of new applications and understanding of the content. Not only is conversational search now supported, the data has become extensible to be used in multiple applications and new product development.
In today’s world, most organizations should be considering implementing an ontology to take full advantage of AI in all forms. Access Innovations can help! Contact us at https://www.accessinn.com/contact-us/.
Taxonomy, Thesaurus, Or Ontology? – Clearning the Confusion, Part One
To read part 1 of this article, please follow the link: Taxonomy, Thesaurus, Or Ontology? – Clearning the Confusion, Part One