RP25 - Recommending scientific literature for revising clinical guidelines with knowledge graphs

The research projects builds on previous work on recommendations for scientific literature for the revision of clinical guidelines using the example of the S3 guideline on melanoma. Here, bibliometrics and modern transformer-based NLP methods were used to recommend literature. The continuation of this topic is dedicated to improving the explainability of literature recommendations based on knowledge graphs. For this purpose, the work refers to existing ontologies/terminologies on melanoma. Based on these, knowledge graphs are created and used in literature recommendation [1]. In recent years, some commercial and free software systems have been developed to support the creation of systematic review papers, e.g. Robotreviewer [2] or DistillerSR. Other current possibilities concern the further development of large language models that are evaluated as knowledge sources (ChatGPT). Here, especially the avoidance of “hallucinations” is relevant, for which the created knowledge graphs are used. It will be investigated to what extent these systems can be meaningfully embedded in and improve the developed literature recommendation. Other foci of the project include conducting an evaluation study with guideline developers and exploring the transferability of the method to other oncology guidelines.

[1] Michalowski, M. Rao, M.; Wilk, S.; Michalowski, W.; Carrier, M. (2023), „Using graph rewriting to operationalize medical knowledge for the revision of concurrently applied clinical practice guidelines“, in Artificial Intelligence in Medicine, Volume 140, 102550, https://doi.org/10.1016/j.artmed.2023.102550.

[2] Marshall IJ, Kuiper J, Banner E, Wallace BC. (2017), „Automating Biomedical Evidence Synthesis: RobotReviewer“. Proc Conf Assoc Comput Linguist Meet. 2017 Jul;2017:7-12. doi: 10.18653/v1/P17-4002