<p>The detection of anomalous behavior has emerged as a critical complement to cybersecurity efforts within online services. This study introduces a novel framework for detecting anomalous user behavior in web applications by leveraging graph-based representations of user activities. Specifically, we explore the application of a geometric feature of graphs called as Forman-Ricci curvature to characterize behavioral graphs constructed from user interactions. The methodology involves representing individual user behaviors as graphs, followed by computing the Forman-Ricci curvature for each graph to capture intrinsic geometric properties. To accommodate graphs of varying sizes, a new distance metric is defined based on graph embeddings, enabling meaningful similarity comparisons across user behavior graphs. Subsequently, clustering algorithms are applied to these similarity measures to assign anomaly scores to users. Empirical evaluation on real-world web application data demonstrates that incorporating Forman-Ricci curvature enhances the detection rate of anomalous behaviors, particularly when combined with diverse embedding and clustering techniques. This approach highlights the potential of geometric graph features in improving the efficacy of anomaly detection systems in web environments.</p>