Mapping Natural Language Questions to Medical Specialties

Many real-world applications handle large amounts of data, which needs to be preprocessed and classified before it can be used. Performing this classification using human agents is usually slow and costly, thus motivating the need for automation. One of the areas that can benefit from an automatized process includes online health-care platforms that connect people to doctors remotely. This type of platform usually offers an asynchronous messaging service, and can receive up to hundreds or thousands of medical questions every day. The questions need to be assigned to the appropriate specialists in a timely manner. This paper offers an automatic solution to this problem. Its purpose is to address the issue of mapping natural language texts to medical specialties. Two solutions are proposed and compared, one based on the Naive Bayes classifier and the other on a Linear classifier implemented using TensorFlow. The former obtains an accuracy of over 95% for 23% of the questions while the latter obtains the same accuracy for 60% of the questions. The results motivate the use of these solutions in real-world applications.