Optimization of sentence classification for insurance applications

This research is carried out on the topic of natural language processing and specifically on word representation on Question Answering tasks. The state of the art in Question Answer task is Google’s Bidirectional Encoder Representations from Transformers (BERT) language model.
Koïos Intelligence is interested in fine-tuning this model for their closed domain artificial intelligence (AI) virtual assistant, targeted at insurance and financial applications. The general objective of this research is to improve the performance of Koïos’s NLP solutions by investigating state of the art sentence embeddings and fine-tuning BERT to the domain of insurance. Specifically, the intern will attempt to improve on BERT by examining how sentences are embedded and the impact of these embeddings on accuracy and performance of the model. The uncertainty of this work lies in whether state of the art sentence embeddings can be improved on in the context of insurance natural language datasets.
The internship project will principally focus on the acceleration and optimization of software developed at Koïos, based on research of state-of-the-art sentence embeddings. The intern will validate and test alternatives for user intent classification in order to improve the system performance.

Faculty Supervisor:

Fabian Bastin

Student:

Partner:

Koïos Intelligence Inc

Discipline:

Business

Sector:

Professional, scientific and technical services; Retail trade

University:

Université de Montréal

Program: