Assessing Consumer Financial Complaints using Text Analytics and Machine Learning
Internationally, overseeing organizations collect tens of thousands of complaints against companies each year, many of which result in actions taken by the companies in question including payouts to the individuals filing the complaints. Given the volume of the complaints, how would an overseeing organization be able to quantitatively assess the data for various trends, including the areas of greatest concern for consumers?
In this presentation, we propose a repeatable model of text analytics techniques to the complaint data from one such organization, the Consumer Financial Protection Bureau of the United States. Specifically, we will explore sentiment and machine learning techniques to model the natural language available in each freeform complaint against a disposition code for the complaint, primarily focusing on whether a company paid out money. This will generate taxonomy in an automated manner, and we will explore methods to structure and visualize the results, showcasing how areas of concern are made available to analysts in a user-friendly environment. Finally, we will discuss the applications of this methodology for overseeing government agencies and financial institutions alike.