FinTech has been at the forefront of using machine learning to streamline business processes. A recent case study involved developing a product classification system for one of our FinTech clients. The system needed to classify over 40,000 product nodes accurately, which presented a challenge due to limited training data. Our team developed alternative training methods and data sourcing mechanisms to overcome this limitation.
To frame the problem in a machine learning context, we conducted a scoped proof of concept (POC) to experiment with different approaches and gain a deeper understanding of the data's quality and representation. Our research included exploring sentence similarity, BERT with dynamic masking, translation services, SVM, Naïve Bays, and XG Booster before ultimately settling on a Deep Learning-based NLP model via multilingual BERT.
Our solution involved a layered Deep Learning Architecture, and we used MLOps tools such as SageMaker, PyTorch, HuggingFace, and Seaborn for continuous training and integration. As a result, our client was able to provide real-time taxonomical classification to end-users at a fraction of the cost of manual human work.