Semi-Supervised Learning for NLP Text Classification

Insurance companies collect huge volumes of text on a daily basis and through multiple channels, which can be used for lots of different analyses, including identifying “cause of death”. It is difficult to overestimate the importance of an insurance company’s need to understand the facts and circumstances surrounding an insured individual’s death. These facts, including the manner and cause of death, along with other data about the decedent, are critical to an insurance company’s ability to measure mortality rates. Considering the huge volume of the data, it is very time-consuming and manual data labelling by human experts is barely possible. The main objective of the proposed research is to develop a semi-supervised model that best suits the unstructured text data. The goal is to develop and validate a generalizable unsupervised deep Natural Language Processing (NLP) model to label the data, identify, and classify “cause of death” from unstructured obituary text.

Faculty Supervisor:

Hadis Karimipour

Student:

Amir Namavar Jahromi

Partner:

Munich Reinsurance Company

Discipline:

Engineering

Sector:

Finance, insurance and business

University:

University of Guelph