Round Corner
Department of Computer and Information Science


Personalized Recommendations at DNB PULS (AI Lab Pitch)


DNB PULS is the new DNB digital product for corporate banking. Our mission is to help our corporate customers in driving their business and stay ahead on the market. We do this through an innovative customer experience delivered by transformative analytics in the DNB PULS app. The DNB PULS Tribe is a team of 28 people, based in Oslo and Bergen and built across two departments of the Corporate Banking branch of DNB – Innovation and Sales (IoS), Strategy and Analysis (S&A) – and IT. The team works according to the agile methodologies, promoting weekly sprints.

In order to achieve its ambition, PULS is continuously working on two core functionalities. These are coming into life through deep synergy between the business and analytics modules. The first functionality is the Future Balance which is an estimate of a corporate’s balance in the next 30 days with daily resolution. The second is a recommender system which generates automated advices based on (among other things) Future Balance, making it easier for the customer to diagnose the status of the company and take action accordingly. Many small companies do not have an advisor and therefore is important to build reliable and reasonable advices. The students involved will have the possibility to improve a recommendation system that generates personalized advices to our big variety of customers. The thesis will be guided by the data science unit of the project. This unit, in cooperation with information architects, is currently setting up an environment on DNBs new public cloud based analytical platform for implementation of model- based recommender systems based on collaborative filtering and matrix factorization. There will be a need for the student to explore and evaluate new implementations of collaborative filtering.

Problem Areas

A current rule-based solution exists, which suggests an advice among a list of around 20. This can be as simple as “if customer is in industry x, suggest a product if it does not already have it”. The PULS team wants to explore machine learning alternatives to rule-based systems, and advices are not limited to up-sales.

Starting from the so-called “cold start problem” in recommendations, the focus of the study will be:

  • Literature review of state of the art recommender systems and related evaluation measures.
  • Identify candidate algorithms to challenge the matrix factorization. Algorithms for proof of concepts (POC) can be found in the field of Deep Learning (for example, Restricted Boltzmann Machines or other Neural Networks) and Reinforcement Learning (for example, Multi-arm Bandit or others “exploit-explore” methods)
  • Implementation and testing of the selected algorithms. Propose measure of evaluations both on accuracy and unexpectedness.

Details can be discussed with supervisor and student.


  • Business Challenges. Interpretability and explainability of the recommender system to the stakeholders and customers.
  • Scientific Challenges. Recent literature showed that the matrix factorization method generates the most accurate recommendations but the least surprising: trade-off accuracy vs. unexpectedness to maximize the customer satisfaction.

Thesis Information

  • 6 months preferred
  • DNB contact: Gianluca Giaquinto (Data Scientist) -

Data and Information Sensitivity

The data will consist of DNBs customer data on corporate customers, and a list of possible advice will be provided.

The department of BM Customer Insight has done already a substantial work with regards to anonymization of sensitive data. An example list of data to be used and level of sensitivity can be
found below:

  • AktorId → Not Sensitive masked internally
  • SelsakpsformKode → Slightly sensitive
  • Omsetning → Sensitive. Can be masked with a ranking function
  • Bruksdato → Slightly Sensitive. Can be shifted if necessary
  • Number of Employees → Medium Sensitivity. Can be masked through a factor function without breaking the numerical proportion
  • Næringskode → Medium Sensitivity. Can be masked with ranking or factor functions

Should the inclusion of more demographical or financial data be necessary we have full control of level of sensitivity and masking rules, and the masking will be performed by DNB.

Data will be made available through DNBs cloud platform.


In this project, the joint interests of sponsor, advisor and student(s) come together. Typically, the project will be based on the problem described by the sponsor AND previous research by Prof. Ole Jakob Mengshoel AND interests of students. If only one or two of these are present, there is no basis for a project.

Please send email(s) to sponsor and/or advisor if you're interested in this project.

Ole Jakob Mengshoel is (more or less) following Keith Downing's selection process for master students - read this:


Ole Jakob Mengshoel Ole Jakob Mengshoel
313 IT-bygget
735 93440 
NTNU logo