A Systematic Literature Review on Graph-Based Models in Credit Risk Assessment

OverviewThis deposit archives the conference presentation materials for a talk delivered at the 4th International Symposium on Big Data and Artificial Intelligence (Hong Kong) on 17 December 2024. The presentation reports a systematic literature review (SLR) on graph-based models for credit risk assessment, with emphasis on applications in banking networks and peer-to-peer (P2P) lending. The talk synthesizes how network/graph methodologies (e.g., centrality-based scoring, community/latent factor models, Graph Neural Networks) complement traditional credit-risk approaches by capturing interdependencies, contagion mechanisms, and dynamic relational information that borrower-level feature models may miss. Contents of this deposit (file-level summary) Conference presentation (PDF): 4th_International_Symposium_on_Big_Data_and_AI_HK_Lennart_Baals.pdf LaTeX source (slides): 4th_Symposium_on_big_data_HKG_Lennart_Baals.tex Figures / assets: Graphics.zip (contains all images referenced by the LaTeX source) What the presentation covers (research scope and key elements) Research questions: how different graph-based model families compare for enhancing credit risk assessment; which approaches are most used; which application settings benefit most; and what challenges/opportunities remain. Systematic review design: literature search in Scopus and Web of Science using graph/network and credit-risk related keywords; structured screening and quality checks; final sample of 78 articles from 1,066 retrieved records. Coding framework: consistent extraction of graph type, task type, application setting (e.g., credit scoring vs. systemic risk), data source, model methodology, metrics, and validation design (including a double-blind coding step). Synthesis of findings: rapid growth of the literature since ~2018; interdisciplinary publication outlets; recurring applications across P2P lending/SME scoring and systemic risk in interbank networks; emerging techniques including GNNs and hypergraph formulations. Challenges and future directions: data access/quality, scalability for large dynamic graphs, and practical integration of graph-based methods with traditional statistical credit-risk models; outlook toward real-time scoring and other network-based credit applications. How to rebuild the slides (reproducibility)To regenerate the PDF from source: Extract Graphics.zip into the directory structure referenced by the LaTeX file (typically a Graphics/ folder). Compile 4th_Symposium_on_big_data_HKG_Lennart_Baals.tex with a standard LaTeX toolchain (e.g., pdflatex or latexmk). The compiled output should match the included PDF (minor differences may occur depending on TeX distribution and package versions). Data and confidentiality noteThis deposit contains presentation documents and figures only. It does not redistribute proprietary or restricted-access datasets. The SLR results are based on published literature and bibliographic screening, as described in the presentation.