Artificial Intelligence (AI) changes how we work and will redefine the relationship between man and machine. It has an incredible growth potential, for both business and personal uses. Info Support, as an innovative solution provider, is working hard to be at the front of AI developments. Therefore research in the domain of Artificial Intelligence is vital and we keep challenging and facilitating our people to do research within this domain.

Research Ambitions

Ambitions are research questions of which the answers will give us the opportunity to determine the future. Are you a master student and would you like to significantly contribute to our thought leadership position in the AI domain? Please reach out to Joop Snijder, Head of Research Artificial Intelligence, for more information about how you can help.

Explainable AI

An Explainable AI (XAI) or Transparent AI is an artificial intelligence (AI) whose actions can be easily understood by humans. It contrasts with the concept of the “black box” in machine learning, meaning the “interpretability” of the workings of complex algorithms, where even their designers cannot explain why the AI arrived at a specific decision. One of our ambitions is to make AI and machine learning more transparent, so customers can understand and trust the models we implement.

Advanced Machine Learning

This research ambition includes research into modern machine learning methods that take advantage of increased complexity to provide improved performance. For example deep learning, reinforcement learning, natural language understanding and computer vision.

Natural Language Processing in Dutch

Natural language processing (NLP) is an area of computer science and artificial intelligence concerned with the interactions between computers and human (natural) languages, in particular how to program computers to process and analyze large amounts of natural language data. There’s a lot of research done for the English language, but no so much for the Dutch language. Our ambition is to expand the research on this topic.

Research Assignments

Bachelor and Master students can choose one of our Research Assignments as a topic for their thesis.

Predicting staffplanning
For retailers in all sectors, personnel planning is a recurring challenge. In order to plan staff, firstly a prognosis is being made. The prognosis is a prediction that says how much production will be asked in the upcoming period. Secondly, based on the prognosis, a prediction will be made of how much staff are required to achieve the requested production. Learn more…

Feature Learning for Machine Learning Applications
Will your thesis be about apllications that use machine learning algorithms? And are you going to do research about what feature learning techniques there are and the application thereof in relation to Info Support Cases? Learn more…

Predicting customer satisfaction based on feedback
Your will research what factors will be of influence on customer satisfaction, based on our data. The available date include structured data(system monitoring and userinformation of applications) and unstructered data (e-mails, incoming phonecalls and messages) Learn more…

Specifying and testing a personality of a chatbot

According to Gartner, by 2020, we communicate more with chat bots than with our partner. This indicates how important chat bots will be in the near future. It turns out that users are humanising chatbots, making it important for chatbot developers to give the chatbot a personality. The question is how to specify and recognize a personality for a chatbot so that you can test the personality.

Answering questions based on documents in Dutch

Employees of companies produce new documents on a daily basis. These documents contain important knowledge. The disclosure of this knowledge is a difficult issue. Not everybody can read anything and search engines only return documents that contain a particular text. For knowledge workers, it seems more convenient to ask questions to a central entity that can respond to the knowledge in documents (such as manuals, terms of employment, procedures, etc.). The research question is therefore: Is it possible to train a question-answer model based on unstructured documents written in Dutch?

Connecting questions and answers in Dutch

More and more companies use internal chat tools, such as Slack. Colleagues start a conversation, ask questions and share knowledge. As you can imagine, answers to asked questions can be useful to other colleagues in the future and that they contain important knowledge. How beautiful would it be if there was a “man-in-the-middle” principle, which, if another similar question arises, could automatically answer the questioner based on historical conversations?

The challenge in such tools is that conversations from different users cross each other which makes it difficult to match responses to the corresponding question. This problem domain is already being investigated for the English language. But Info Support is mainly active in the Dutch language field and wants to investigate whether it is possible to link questions and answers in Dutch.

Scoring the confidence of an answer given by a FAQ model

Info Support develops various chatbots for companies to reduce the traffic to their customer support department by answering frequently asked questions. For this the chatbot uses a so-called FAQ model built as deep neural network that provides a probability score for all possible answers and picks the one with the highest probability.

This however doesn’t always mean we get a right answer. The probability scores fluctuate wildly, because of influences such as writing style of the customer, the length of the question and the size of the vocabulary used to train the model.

We want to expand our FAQ model with a score that tells us how confident we are the answer given will be useful for the customer. We want to use this confidence score as a way to decide when to transfer a customer to a real customer support agent so that the customer has the best chat experience possible.

Research if such a score could be calculated and define a method to calculate the confidence score. Finally, a test method to validate if the confidence score helps to streamline the chatbot conversation.

Virtual Pair Programmer which predicts the next lines of code

Pair programming is an agile software development technique in which two programmers work together at one workstation. One, the driver, writes code while the other, the observer or navigator, reviews each line of code as it is typed in. This helps to produce better code. At Info Support we have a lot of historic data about code. The question is: is it possible to have a virtual pair programmer which helps the developer to code by predicting the next lines of code.

Virtual Pair Programmer that predicts if not written tests would fail for code parts

Pair programming is an agile software development technique in which two programmers work together at one workstation. One, the driver, writes code while the other, the observer or navigator, reviews each line of code as it is typed in. This helps to produce better code. At Info Support we have a lot of historic data about code and the tests and results belonging to that code. If a virtual pair programmer can predict that the code won’t work (if you would write tests for it), it will increase the productivity of the programmers and make their lives easier.

Virtual Unit Tester that predicts Unit Tests for the programmer

Unit Testing is a level of software testing where individual units/ components of a software are tested. The purpose is to validate that each unit of the software performs as designed. In theory programmers write unit tests before they code the unit. In practice the programmer writes the unit tests afterwards. This is tedious task.

Info Support has a large historical data set containing source and accompanied unit tests. The question is: is it possible to predict unit test for source code written in C#

Artificial Intelligence Theses

In addition to our publications, Info Support Research also has a myriad of relevant theses.

Using Discretization and Resampling for Privacy Preserving Data Analysis: An experimental evaluation  

Master of Business Informatics. Utrecht University. November 2018

Sven Pors

Method Call Argument Completion using Deep Neural Regression

Master Software Engineering. University of Amsterdam. August 2018

Terry van Walen

Unit test generation using machine learning

Master Software Engineering. University of Amsterdam. August 2018

Laurence Saes

Quantifying Chatbot Performance by using Data Analytics

Master in Business Informatics. Utrecht University. July 2018

Cas Jungerius

Code Completion with Recurrent Neural Networks

Master Software Engineering. Utrecht of Amsterdam. Spring 2018

Erik van Scharrenburg

Chatbot Personality and Customer Satisfaction

Bachelor Information sciences. Utrecht University. February 2018

Hayco de Haan

Specifying and Testing Conversational User Interfaces

Bachelor Information sciences. Utrecht University. July 2017

Jasper Berends

Building a Data-Driven Search Engine Spelling Corrector

Master of Science Thesis. University of Twente

Bas Niesink & Stefanie Stevens

Automated Taxonomy Expansion and Tag Recommendation in a Knowledge Management System

Master of Science Thesis. University Utrecht.

Maarten van Duren

Are you interested in working in this research area?


Don’t hesitate to reach out!

Head of Research
Artificial Intelligence


Joop Snijder