|
Abstract:
|
We report the applicability of the practical use of text classification and clustering methods to improve the efficiency and effectiveness in evaluating the unstructured data representing the voice of the customer. Medical voice of customer (mVoC) is feedback on the medicines collected from various customer groups including patients, payers, providers, and policy makers in the health care environment. With the ever-growing amount of such information, an automatic medical text classification and clustering tool is becoming increasingly important. To better understand the medical and scientific needs, and the concerns that drive customer's decision-making and behaviors, we develop a tool called "Medical Insight Explorer" to assist the work. The software utilizes multinomial Naive Bayes method to classify the mVoC data into various domains, and under each domain, non-negative matrix factorization (NMF) is employed to identify trends and topics in the data. The experiment results show how we were able to increase performance through the use of these algorithms over other underlying methods.
|
