지역밀착 거래 플랫폼의 제품판매정보 텍스트 분류를 활용한 개인정보 예측에 관한 연구

This study aims to develop a model for diagnosing the risk of personal information exposure among users of hyper-local product transaction services. The research focuses on predicting and classifying gender information and marital status through text analysis by securing a dataset of sales posts. To achieve this, the sales post dataset was collected, and labeling for gender and marital status was performed. Latent Dirichlet Allocation (LDA) topic analysis was then used to classify the types of sales posts and identify their characteristics. Finally, a Naive Bayes model was developed to diagnose the risk of gender and marital information exposure. The data was scraped from posts on a hyper-local transaction service platform and analyzed using various packages in the R programming language for web scraping, topic analysis, and Naive Bayes analysis. The LDA topic analysis revealed that certain words frequently appeared in sales posts based on product type, allowing for the estimation of the seller's gender and marital status. Using this information, a Naive Bayes model was developed to predict the risk of exposure of the seller's gender and marital status, with a gender prediction accuracy of 74.6% and a marital status prediction accuracy of 95%. The results of this study are expected to contribute to safer transactions by identifying product types and words with high personal information exposure risk.