Explainability in Recommender Systems - Interview with Behnoush Abdollahi

Behnoush Abdollahi is a Ph.D. research assistant at the Knowledge Discovery and Web Mining Lab, University of Louisville. Her research focuses on personalization systems and techniques incorporating large amounts of data available on the internet — for example : user activity on social networks and user reviews for items on e-commerce websites.

In addition to working on her Ph.D., in 2015 Behnoush joined start-up company Gen-9 as a Machine Learning Software Engineer Intern, helping to build multiple data products and applications (including mobile) to study user patterns of varying time frames using wearables. Additionally, working as a graduate research assistant at the BioImaging Lab, University of Louisville, wokring on machine learning techniques for image segmentation of lung/chest regions in medical CT images using Parallel Programming and GPU computing.

We’re looking forward to hearing more during Abdollahi’s talk at AI With the Best online conference April 29–30th, and are pleased to have had a chance to interview the recommender researcher ahead of this event.

Q. What personally motivated you to begin your work in machine intelligence?

The challenge of thinking and the use of logic to solve a problem always attract me to mathematics and theory. Machine learning, on the other hand, motivates me to practically apply my knowledge to build algorithms for solving specific problems using data, and this is not only interesting, but very useful. Machine learning has changed my mindset about approaching a problem by providing a powerful set of tools for data-driven problem solving. Also, with the fast growing of data and computing powerthe opportunity to build new applications and have impact on the lives of people have increased, at both individual and social scales. This is very exciting and motivating to me and I look forward to many amazing achievements in this field.

Q. Tell us about your latest research project?

While building an accurate machine learning models seems like the ultimate and ideal goal, the minimum and urgent criterion, that machine learning models should satisfy, is transparency. Therefore, designing interpretable intelligent systems, that facilitate conveying the reasoning behind the results, is of great importance. This is the scope of my PhD dissertation research. More specifically, in my research I chose to work in the context of recommender systems. This is because recommender systems are a prominent example of a AI model that interacts directly with humans (users), in contrast to many other traditional (e.g. medical) decision making systems that interact with experts (e.g. healthcare providers).

Q. How has the field of recommender systems changed since you began working in it?

In recent year, the focus of recommender systems has largely changed towards context-aware models and personalization-based methods using human computer interaction and cognitive science techniques. Deep models, have also been shown to give promising results mostly in content-based recommender systems and hybrid models. In addition, the ability of intelligent systems to explain their decisions and actions in many systems where the end user is not an expert, such as recommender systems, is very limited. The need to design and build explainable AI models, including recommender systems is increasing rapidly.

Q. What advice would you give to budding AI developers?

I think gaining a strong background in basics of applied math, optimization, and statistics is very important for beginners, in addition to learning about the machine learning and statistics techniques and toolkits. After becoming familiar and confident with the process, practicing on datasets and diving deeper into the details of algorithms are keys to becoming expert.

Q. Are you excited about speaking at AI With The Best? What made you want to be a speaker?

Yes! I like that AI With The Best is an online conference and brings developers and researchers from around the world together. This increases the opportunity to attend technical conferences, mainly for students and general audience. I can’t wait to participate in the AI With The Best conference.

Thank you Behnoush!

You can ask Behnoush your own questions, and learn more in her talk about using Explainability in Recommender Systems at our upcoming AI With The Best, Online Developer Conference 29–30th April.