Driving Innovation Together
To achieve the ambitious goals of the KGELL Action, our collaborative efforts are structured into six specialized Working Groups (WGs). Each group brings together academic experts and industry practitioners to tackle specific challenges at the intersection of Knowledge Graphs and Large Language Models.
WG1: Augmenting LLMs with KGs
The objective of this WG is to explore strategies for using Knowledge Graphs to improve LLMs, making them more accurate and trustworthy by injecting knowledge from KGs into LLMs
Key Tasks
- Creating benchmarks to measure factfulness, hallucinations, and bias in LLMs.
- Exploring fine-tuning methods using factual knowledge from KGs.
- Utilizing KGs as plugins (like Retrieval Augmented Generation) to query factual data.
- Developing post-processing methods to verify and rectify LLM outputs against KGs.
WG2: Domain-specific tasks of LLMs based on KGs
This WG focuses on analyzing various tasks that can be effectively tackled using LLMs integrated with KGs across different domains, creating data and algorithmic resources to address specific challenges.
Key Tasks
- Systematically exploring downstream tasks within specific domains.
- Mining and curating extensive datasets from scientific papers, patents, news, finance, and medical reports.
- Training LLMs for optimal performance within these specified domains.
- Organizing challenges and competitions to promote innovation and interdisciplinary collaboration.
WG3: KG Construction assisted by LLMs
The primary objective is to analyze and develop new methodologies for generating KGs, focusing on semi-automated information extraction methods and creating an open library of methods and resources.
Key Tasks
- Developing best practices for ontology engineering and schema reuse.
- Using LLMs for information extraction (named entity recognition and relationship extraction).
- Addressing challenges related to publishing, storing, and querying KGs.
- Interlinking new KGs with existing public networks (e.g., Wikidata, DBpedia).
WG4: Multilinguality for KGs and LLMs
This WG aims to improve how KGs can capture, connect, and utilize information across various languages (particularly low-resource languages) to overcome language barriers and enhance global information exchange.
WG5: Bias and Ethics
This WG will focus on advancing methods that promote fairness, transparency, and accountability in the construction, maintenance, and use of KGs.
Key Tasks
- Detecting and mitigating unfair biases related to ethnicity, gender, and race.
- Improving the interpretability and explainability of KG construction processes.
- Providing ethics training and engaging with diverse stakeholders.
- Implementing domain-specific debiasing (e.g., avoiding biased generalizations in healthcare models).
WG6: Evaluation and Validation Frameworks
WG6 will focus on creating standardized benchmarks and evaluation methodologies to systematically assess LLM-KG integrations, testing their practical effectiveness, scalability, and ethical implications.
Key Tasks
- Designing specific benchmarks in collaboration with all other Working Groups.
- Establishing standardized metrics for accuracy, consistency, knowledge retention, and cost-benefit analysis.
- Creating validation protocols to test integrations in real-world, high-stakes environments like healthcare and finance.
