Hybrid Recommendation System for Financial Institution
Name
Ivan Slobozhan
Abstract
Recommendation systems are often used by companies to target customers with personalized offers. This, in turn, helps to increase the revenue from marketing campaigns and improve customers' experience. Recommendation systems are commonly used in e-commerce sites (Amazon, E-bay) and entertainment platforms (Spotify, Youtube). However, their use has not yet been broadly explored in the financial sector.
In this thesis, we propose and evaluate a hybrid recommendation system algorithm to generate personalized offers for customers of a bank. The recommendation system algorithm uses implicit information about customers' transactions with companies in order to recommend companies that customers have not recently visited, and that they might wish to visit in the near future. The algorithm is shown to be robust enough to overcome the cold start problem, which in our case is the lack of data from customers with a small transaction history. The algorithm evaluated using real datasets (customer's transactions), which are provided by a major North-European bank. Compared to a random recommendation model, which is presently in use by the bank for their marketing campaigns, our recommendation algorithm has a maximum lift value against random targeting of 416 and minimum 55 which shows the effectiveness of our approach.
In this thesis, we propose and evaluate a hybrid recommendation system algorithm to generate personalized offers for customers of a bank. The recommendation system algorithm uses implicit information about customers' transactions with companies in order to recommend companies that customers have not recently visited, and that they might wish to visit in the near future. The algorithm is shown to be robust enough to overcome the cold start problem, which in our case is the lack of data from customers with a small transaction history. The algorithm evaluated using real datasets (customer's transactions), which are provided by a major North-European bank. Compared to a random recommendation model, which is presently in use by the bank for their marketing campaigns, our recommendation algorithm has a maximum lift value against random targeting of 416 and minimum 55 which shows the effectiveness of our approach.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Rajesh Sharma
Defence year
2018