<![CDATA[This study aims to analyze topics in banking customer complaint data using the Latent Dirichlet Allocation (LDA) method, enhanced with parameter tuning via Grid Search. The dataset is sourced from ConsumerFinance.gov, containing a total of 6.3 million complaint entries from 2011 to 2024, with 50% of the data used to maintain representation and simplify analysis. In this analysis, the LDA method is employed to identify hidden topics, while Grid Search enhances model coherence. The results indicate that customer complaints can be categorized into 10 main topics, including complaint report issues (25.67%), payment errors (18.10%), data authorization (12.20%), and credit policy (10.77%). Parameter optimization successfully improved the model's coherence score from 0.49 to 0.56, reflecting an enhancement in topic clustering quality. A comparison between standard LDA and LDA with Grid Search reveals that the optimization method yields a higher average coherence score (0.52 vs. 0.42). This study provides insights into common complaints received by banks and key terms such as "report," "authorization," and "investigation," which can assist banks in better understanding and addressing customer complaints more effectively.]]>
Copyrights © 2025
