Application of Computer Graphics Visualization Techniques on LDA Topic Modeling Results of Banking Customer Complaint Data

Abstract

Topic modeling using Latent Dirichlet Allocation (LDA) produces abstract results in the form of word distributions per topic that are difficult to interpret without proper visualization. This study applies computer graphics visualization techniques—specifically Word Cloud rendering using the Flood Fill algorithm and Archimedean Spiral Placement—to represent the results of LDA topic modeling on banking customer complaint data from ConsumerFinance.gov (6.3 million entries, 2011–2024). Ten major complaint topics previously identified with an optimized coherence score of 0.56 were visualized through Word Clouds, where word size is proportional to weight within each topic and color coding distinguishes topic categories. The Flood Fill algorithm is applied for region-based coloring of word areas, while Spiral Placement determines non-overlapping word positioning. Results show that the visualization effectively communicates topic prominence and keyword relationships that are less apparent in tabular form. The Word Cloud representations successfully highlight dominant complaint themes such as credit reporting (25.67%), payment errors (18.10%), and data authorization (12.20%). This approach demonstrates that computer graphics techniques can meaningfully enhance the interpretability of NLP outputs in data-driven decision-making contexts.