Time-Sensitive Embedding for Understanding Customer Navigational Behavior in Mobile Banking

Hakvar H., Cavuldak C., Söyler O., Karadayı Y., AKTAŞ M. S.

International Conference on Computing, Intelligence and Data Analytics, ICCIDA 2022, Kocaeli, Turkey, 16 - 17 September 2022, vol.643 LNNS, pp.257-270 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 643 LNNS
  • Doi Number: 10.1007/978-3-031-27099-4_20
  • City: Kocaeli
  • Country: Turkey
  • Page Numbers: pp.257-270
  • Keywords: Aging based embedding, Customer behavior, Customer embedding, Graph base embedding, Representation learning
  • Yıldız Technical University Affiliated: Yes


The availability of digital products, mobile, and the internet has become very widespread in the financial world, which provides a lot of information regarding customer financial habits. Purchasing habits of customers and their tendency to buy a specific product can be determined by the footprints left by customers in digital banking applications. However, tendency based on visit frequency and time-dependency that might provide valuable data for representing customer behavior are generally missed out by generic embedding approaches. We commit this problem by suggesting a customer embedding framework to leverage the time-sensitive digital footprint of mobile banking customers to gain benefit in prediction scenarios. The proposed framework utilizes only the digital footprints of customers generated on a mobile banking application. It helps us to represent customers who don’t have much contextual history in the banking environment and accurately predict their investment tendency. We test customer embedding vectors generated by the proposed framework using real-world digital footprints of mobile banking customers in a prediction scenario in which the intention of customers towards a financial product is predicted. The proposed customer embedding framework has shown better performance over the plain usage of state-of-the-art embedding approaches Word2Vec and DeepWalk.