Abstract: Many of us interact with recommender systems in one way or another in our daily life. These systems recommend items of interest to users. For example, YouTube recommends videos, while Spotify recommends audio recordings. In this presentation, I will talk about a subcategory of recommender systems, which can suggest items that exhibit a particular feature with specified intensity. For example, a user can specify that they want to be recommended movies that are more about time travel and less about cars compared to their current movie recommendations. These systems require rich metadata regarding items, such as information on the degree, with which a feature applies to an item. For example, this could be a value between 0 and 1, which indicates how much drama applies to the movie Titanic (1997). In this presentation, I am going to talk about how we are building this kind of dataset. I will cover two topics: surveying users regarding item features and predicting user answers with machine learning. As this research is in its early phase, I am going to present preliminary results of our experiments and mention challenges we face while working on the problem. The goals of this presentation are to (1) let FCAI members know what our team is currently working on, (2) receive feedback on our ongoing research and (3) look for possible collaborations within FCAI.

Speakers:  Denis Kotkov

Denis is a postdoctoral researcher at the University of Helsinki working in the ESP group (https://glowacka.org/) on recommender systems and exploratory search. He completed my PhD at the University of Jyväskylä in 2018. The topic of his dissertation was “Serendipity in Recommender Systems”. His homepage: http://denis.kotkov.me/

Affiliation:  University of Helsinki

Place of Seminar:  Zoom