Description
In ecology, systematic reviews and meta-analyses are important methods to synthesize scientific evidence. They allow to qualitatively and quantitatively integrate large bodies of literature to investigate cross-study research questions. Ecological evidence is described by numerous interrelated and complementary entities, many of which are associated with spatial and temporal information. Those entities and their relationships can be represented as spatiotemporal knowledge graphs (STKGs) enabling the analysis of complex ecological interdependencies, e.g., based on the identification of patterns and connections across studies.
The construction of knowledge graphs is increasingly being supplemented by large language models (LLMs). However, the spatiotemporal and complex nature of some domains, such as ecology, has received limited attention so far. Automated STKG construction methods could support publication analyses and synthesis by enabling efficient extraction, representation and integration of ecological knowledge. Despite this potential, LLM-generated knowledge graphs can be unreliable and insufficiently evaluated. Therefore, close interdisciplinary collaboration with domain experts is essential to integrate their knowledge and develop new benchmarks for assessing the quality and reliability of automatically constructed STKGs.
We work with researchers in the “AgriRestore” project, that aims to investigate key indicators for more resilient ecosystems in the field of ecological restoration by evaluating measures for ecosystem and landscape restoration in agricultural landscapes. The project’s meta-analyses, ecological studies and the domain-specific knowledge of the collaborators provide a valuable framework for developing, evaluating and interpreting generated STKGs. With our presentation, we aim to share our current progress on how to use LLMs to derive STKGs from publications and effectively support literature synthesis methods in the field of ecology.
| Topic | Topic 4: Can knowledge graphs and AI-based synthesis improve ecological synthesis? |
|---|