Parallelized Latent Dirichlet Allocation for Medical Fraud Detection
This paper seeks to analyze topic modeling software that assists investigators search for potential fraudulent Medicare billing activities. Furthermore, it will document the research done to parallelize this software in order to reduce its computational processing time. This paper also seeks to study questions such as “do medical providers of a specific type have similar billing patterns across states and regions?” and “can semantic analysis be used to identify both common and outlier billing patterns for specific providers?” According to the Blue Cross, medical fraud costs the United States government $68 billion every year. This project focused on improving the speed and usability of the existing software to combat this problem.
Latent Dirichlet Allocation, LDA, topic modeling, topic clustering, CUDA, medical fraud detection, medicare fraud, parallel programming, Honors College
Ready, J. C. (2020). Parallelized latent Dirichlet allocation for medical fraud detection (Unpublished thesis). Texas State University, San Marcos, Texas.