ConferenceCall 2024 03 06

Session	LLMs, Ontologies and KGs
Duration	1 hour
Date/Time	6 Mar 2024 17:00 GMT
	9:00am PST/12:00pm EST
	5:00pm GMT/6:00pm CET
Convener	Gary Berg-Cross

Ontology Summit 2024 LLMs, Ontologies and KGs

Agenda

• 12:00 - 12:07 Gary Berg-Cross Summary of some issues from Session 1 and some foundational issues for LLMs vs Cognitive/Symbolic systems Slides
• 12:07 - 12:37 Hamed Babaei Giglou Exploring LLMs for Ontology: Ontology Learning and Ontology Matching

Abstract: Large Language Models for Ontology Learning (LLMs4OL) framework utilizes Large Language Models (LLMs) for Ontology Learning (OL). LLMs have shown significant advancements in natural language processing, demonstrating their ability to capture complex language patterns in different knowledge domains. Within the LLMs4OL framework, we investigate the "Can LLMs effectively apply their language pattern capturing capability to OL, which involves automatically extracting and structuring knowledge from natural language text?"

To explore LLMs in the area of OL, the first challenge is the formulation of the work and where to use LLM to be suitable for the given task. We conduct a comprehensive analysis using the zero-shot prompting method to evaluate nine different LLM model families for three main OL tasks (where our formulation to use LLMs comes in): term typing, taxonomy discovery, and extraction of non-taxonomic relations. Additionally, the evaluations encompass diverse genres of ontological knowledge, including lexicosemantic knowledge in WordNet, geographical knowledge in GeoNames, and medical knowledge in UMLS.

The obtained empirical results show that foundational LLMs are not sufficiently suitable for ontology construction that entails a high degree of reasoning skills and domain expertise. Nevertheless, when effectively fine-tuned they just might work as suitable assistants, alleviating the knowledge acquisition bottleneck, for ontology construction.

Within this presentation, the main focus will be given to experimented LLM variants and results analysis within the OL task. The whole project is released to the community with detailed documentation here: https://github.com/HamedBabaei/LLMs4OL

Bio: Hamed Babaei Giglou is a researcher at TIB and is currently, involved in the Neural-Symbolic SCholarly InnovatioN EXTraction (SciNEXT) project in collaboration with Open Research Knowledge Graph (ORKG) project at TIB -- German National Library of Science and Technology. Hamed got a bachelor's and Master's degree in computer science and worked as an NLP Researcher for more than 3 years in the industry before joining the TIB as a PhD candidate. Currently, he is pursuing a PhD in "Computer Science: NLP and Semantic Web Technologies" under the supervision of Dr. Jennifer D'Souza and Prof. Dr. Sören Auer. His current research focuses on employing LLM in various ontology tasks, such as Ontology Learning and Ontology Matching.

• 12:37 - 13:00 Discussion

Conference Call Information

• Date: Wednesday, 6 March 2024
• Start Time: 9:00am PST / 12:00pm EST / 6:00pm CET / 5:00pm GMT / 1700 UTC
  • ref: World Clock
• Expected Call Duration: 1 hour
• Video Conference URL: https://bit.ly/48lM0Ik
  • Conference ID: 876 3045 3240
  • Passcode: 464312

The unabbreviated URL is: https://us02web.zoom.us/j/87630453240?pwd=YVYvZHRpelVqSkM5QlJ4aGJrbmZzQT09

Foundational Ontologies

[12:21] ToddSchneider: How can existing foundational ontologies be incorporated into this work?

Biomedical

[12:33] DrRavi Sharma : Why NCi related results are lower, is it hat the information is highly correlated in medicine

Reasoning

[12:35] ToddSchneider: Was a reasoner used to test the (logical) consistency of the resulting artifacts?

[12:41] ToddSchneider: How could the results of a reasoner be incorporated in the ‘process’? That is, how could a reasoner be used as part of the ‘learning’ process?

[12:47] PhilJackson : can the system learn ontologies of verbs (actions) or adjectives (properties), as well as ontologies of nouns (classes)?

[12:51] BobbinTeegarden: Doesn't reasoning imply a logical structure, and is there a 'logical structure' assumed in a neural net process?

[12:56] ToddSchneider: Logical reasoning is applied to a collection of logical and non-logical symbols, along with rules of inference (on those symbols).

Track D: Finance Panel

[12:56] MikeBennett : Finance is still looking for speakers

Next Meetings

	Session
ConferenceCall 2023 10 04	Overview
ConferenceCall 2023 10 11	Setting the stage
ConferenceCall 2023 10 18	A look across the industry, Part 1
... further results