Loading...
Paper Type
Complete
Description
Traditional recommender systems are single objectives, recommending the most relevant items. Unfortunately, this can harm the user experience and limit user interaction. The Calibrated Recommendation System is a multi-objective recommender system that desires to provide relevant and calibrated items simultaneously. The calibration objective seeks to generate a recommendation list that observes the genre proportions of the items in the user's preference. The kernel idea is inserting items that follow these proportions. However, there are other objectives that a recommender system can achieve besides calibration. Thus, this study aims to analyze the effect of the calibration over other objectives such as novelty, coverage, personalization, unexpectedness, and serendipity. We implement four state-of-the-art proposals and compare them against the baseline. Two datasets were used to run the experiment. We discover that calibration implies more items covered, more personalized recommended lists, and more unexpected items, producing serendipity.
Paper Number
1316
Recommended Citation
Silva, Diego Corrêa and Durao, Frederico Araujo, "How Novel and Unexpected the Calibrated Recommendations Are?" (2023). AMCIS 2023 Proceedings. 6.
https://aisel.aisnet.org/amcis2023/sig_odis/sig_odis/6
How Novel and Unexpected the Calibrated Recommendations Are?
Traditional recommender systems are single objectives, recommending the most relevant items. Unfortunately, this can harm the user experience and limit user interaction. The Calibrated Recommendation System is a multi-objective recommender system that desires to provide relevant and calibrated items simultaneously. The calibration objective seeks to generate a recommendation list that observes the genre proportions of the items in the user's preference. The kernel idea is inserting items that follow these proportions. However, there are other objectives that a recommender system can achieve besides calibration. Thus, this study aims to analyze the effect of the calibration over other objectives such as novelty, coverage, personalization, unexpectedness, and serendipity. We implement four state-of-the-art proposals and compare them against the baseline. Two datasets were used to run the experiment. We discover that calibration implies more items covered, more personalized recommended lists, and more unexpected items, producing serendipity.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIG ODIS