A Study on Analytical Models for Customer Inquiry Data Recorded in Natural Language Format

Abstract

As efforts to utilize various types of data accumulated by companies continue to advance, customer complaints are increasingly being recognized as a valuable source of information for marketing activities. Although the volume of such data makes manual analysis difficult, the application of machine learning offers a promising solution. By analyzing complaint data, it is expected that companies will be able to identify current issues and customer needs, leading to improvements in products and services, as well as the development of more effective employee training programs. Customer complaint data is typically collected through various channels. It generally consists of both structured text data, which is recorded with predefined tags, and unstructured text, which is written freely by individuals. However, the data often varies in structure and expression depending on the author, and the level of detail is not consistent across all entries. In addition, complaint data tends to include a large amount of supplementary information unrelated to the actual complaint, which becomes noise in the analysis. While various machine learning-based review analysis methods have been proposed, applying them directly to such inconsistent and noisy text data often makes it difficult to extract meaningful insights. To address this, the present study targets large-scale customer complaint data that is inconsistent in format and includes significant noise, with the aim of developing an analysis method that enables the extraction of useful insights. The present approach first extracts meaningful information from the complaint data and then applies conventional analytical methods. Specifically, a method is proposed that combines pattern matching with context-based extraction techniques to isolate text segments suitable for analysis. By applying the proposed framework, document classification techniques such as labeling and clustering can be effectively utilized in accordance with complaint content. This enables not only the identification of overall trends in the data but also a deeper understanding of current issues and customer needs. Finally, the effectiveness of the proposed method is demonstrated by applying it to real-world complaint data.