ABSTRACT
This chapter presents four case studies, namely, risk assessment, fraud detection, sentiment analysis, and life status estimation, based on both real and simulated datasets and events. The implementations make use of all three tools, iDAS, E5, and aText, described in the last section. The risk assessment case study takes a hybrid model-based/machine learning approach, whereas the life status estimation case study, which is temporal in nature, is purely model-based. Many generative/discriminative models and supervised/unsupervised clustering techniques presented in this book are suitable for detecting various types of fraud, and we provide overall guidelines instead of one detailed modeling approach. The sentiment analysis case study makes use of an unstructured textual corpus of customer surveys.
