Whether shopping on Amazon or watching Netflix, most technology platforms use recommendation algorithms to determine the specific content pushed to certain users. From the user’s perspective, these recommendations can sometimes feel perfect; other times, they can simply feel repetitive and digitally isolating.
As recommendation systems have become an increasingly popular aspect of technological platforms, researcher Konstantin Bauman has set out on a mission to augment and complement current practices.
Bauman, assistant professor at the Fox School’s Department of Management Information Systems, was recently awarded a grant of $1,500 by Temple University Faculty Senate’s Research Programs and Policies Committee (RPPC) to support his research on recommender systems.
Bauman’s project, Empowering Recommender Systems: Helping Users Explore their Own Preferences, proposes a new approach for recommender systems, called Empowering Recommender Systems (EmpRS).
“Recommender systems are an increasingly valuable decision tool for consumers in online settings,” says Bauman. “Traditional recommender systems methods aim to identify a set of items that the user would potentially like. Their ultimate goal is to learn user preferences and automatically select the best matching items for recommendations.”
Traditional recommender systems aren’t without flaws, though. According to Bauman, these systems can raise critical issues, like filter bubbles, in which users are only familiar with a limited subset of items and never gain exposure to items away from those subsets.
“For example, in a music application, a user may like and often listen to artists in the blues category. Traditional recommendation approaches would recommend tracks within this and related music categories, such as bluegrass, and would not steer very far away from them,” explains Bauman. “Therefore, this user may not be aware of the existence of categorically distant genres, such as various folk music traditions, which the user may enjoy.”
To avoid the issues of traditional recommender systems, the EmpRS concept aims to better help users establish their preferences, explore the domain and enhance their level of confidence in their preferences.
According to Bauman, the goal of the EmpRS recommender approach is to empower users with preference-related self-knowledge that allows them to better explore and search for the most desirable items on their own.
In his proposed study, participants will be divided into three experimental groups and asked to listen to various classical musical tracks, rate them and then share their confidence in those ratings. Then, participants will be provided recommendations using either a baseline or EmpRS approach. After, participants will be asked to create playlists of music they would like to listen to.
“The aim of the study is to show that participants who interact with our empowering recommendation system would be able to create more diverse playlists in shorter periods of time than participants receiving other traditional types of taste-based recommendations,” says Bauman.
The proposed EmpRS recommender system could serve as a tool to complement and augment existing recommender systems, according to Bauman.
Having worked on recommender systems models for around ten years, Bauman says he hopes the methods he develops can “help real users to get better and more personalized service in different areas, such as music listening and education.”
The RPPC grant, which Bauman intends to use to pay the study’s participants, is just the beginning of his EmpRS journey.