Dimitris Paraschakis, doctoral student in Computer Science, is defending his thesis Algorithmic and Ethical Aspects of Recommender Systems in e-Commerce.
Open to the Public on Friday 16th March, Malmö University (Niagara House NIB: 0E15) at 13.15 - 15.00
Apptus has had the pleasure of working with Malmö University over a number of years. On Friday 16th March the results of one such collaboration will be presented to the public by Dimitris Paraschakis.
Dimitris will be defending his licentiate degree by holding a seminar on his area of expertise, 'Recommendation Algorithms'. Developed in conjunction with Apptus, the seminar will be held at Malmö University (Niagara House NIB: 0E15) at 13.15.
During the seminar he will defend his articles and discuss the following areas:
- How do simple algorithms compare to popular algorithms in the world of research?
Comparing the 'those who bought also bought’ type algorithms with 'Matrix Factorisation Models', Dimitris will outline how, in practice, the latter are outclassed by the simpler algorithms. He proves the point through a methodical review of how the algorithms perform on real data in natural conditions.
- How do you address the “cold start” problem in recommendation algorithm theory?
“Cold start” is the problem associated with presenting relevant products when behavioural data is missing (e.g. for new products or for products that for other reasons are in the long tail).
To address this issue, Apptus has developed an advanced algorithm that uses Thompson sampling and reinforcement learning. This algorithm is known as 'Synergy' at Apptus. Paraschakis has analysed the algorithm’s performance in handling large amounts of data and demonstrated how difficult it is to beat the Thompson sampling method even when we actually have some information about the future and, therefore, the ability to ’cheat’ the learning process. The findings are a confirmation of the algorithm's stability and efficiency.
- Recommendation Algorithms and Ethics
Dimitris is also interested in ethics within recommendation algorithms. These algorithms utilise people's behaviour to streamline the recommendation system, which means that developers will encounter ethical issues that should be dealt with in a correct and responsible manner. He has started research in this area where very little is known, which was demonstrated by the recent debate in the wake of the US Presidential election in 2016. As such it is considered an open topic that is ripe for exploration - and Dimitris is well positioned to lead the race in its development.