TABLE 5.

Examples of types of inquiries as well as the outcome of both heuristics and ML approaches have been presented. Due to data sensitivity, product names have been omitted and original inquiries have been slightly manipulated.

Subcategory type	Sample inquiry	Heuristic approach	Weakly supervised approach
Real world evidence	A customer requested information about clinical trial study data for a patient population of women having a pre-existing condition	For this type of inquiry, the system falsely categorized the inquiry as specific clinical study or trial result	For this type of inquiry, the system was able to predict the accurate category even when the information was not explicitly mentioned in the text
Efficacy	An HCP (healthcare professional) requested information if a patient can be treated with one of our products after undergoing a treatment regimen with other product	In this example, the system was able to correctly identify the category as efficacy since our ruleset was able to capture such patterns of expressions in the data	The system was not able to categorize this inquiry probably due to a smaller number of training examples for the category efficacy
Publication request	A customer requested for scientific references in terms of publication or poster explaining efficacy of one of our products	For this type of inquiries where explicit request for a scientific resource was made, both heuristics and machine learning algorithms were able to categorize the inquiries accurately