Recommender systems for product sales in the banking industry
Jorge Peña · eSource (Dublin Business School) · 2021
Recommender systems can be found nowadays in a variety of different domains and industries, its popularity depends on the proven exponential enhancement of streaming services and e-commerce businesses on recent times. However, most of the recommender systems rely on the ratings given by users to the items. The major challenge for using recommender systems in the banking industry is the absence of ratings. It is very unlikely that a bank institution will ask customers to rate their products. To solve this problem, the research implemented a rating algorithm that gives a weight depending on the number of transactions and product usage from every customer of the dataset. Customer segmentation was used for targeting the customers that were more prone to open and accept an investment from the banking institution. PyTorch helped to develop a Neural Collaborative Filtering model that was capable to predict the interaction between the customer and future banking products. Finally, PySpark was used for the development of an ALS recommender system that generated banking products recommendations to the customers. This research contributes to future studies by experimenting with multiple recommender systems applied to the banking industry that will increase the value of customers and support decision making on implementing customer marketing campaigns.