AskCFIA webpage is one point of entry for industry to ask questions (open-ended, unstructured texts in English or French) about regulatory requirements.
These questions are triaged manually into three different levels, based on their complexity, before they are answered.
The primary goal of this project was to create a machine learning algorithm (Using R Programming) that could automatically and accurately predict the “level” of the question based on certain characteristics.
The current pilot project has produced a model that predicts whether a given question is level 1 or not (binary outcome), with accuracy of about 80%. Descriptive analysis of the text corpus were also conducted as part of this pilot.
Further development of the model is being planned to increase the predictive accuracy and explore automated response-generation.