Detecting polarization in ratings: An automated pipeline and a preliminary quantification on several benchmark data sets

• DOI: 10.1109/bigdata.2017.8258231
• Publication Date: 2018-01-15T22:47:28Z
• Authors: Mahsa Badami; Olfa Nasraoui; Welong Sun; Patrick Shafto
• Keywords: Benchmark (surveying)

Personalized recommender systems are becoming increasingly relevant and important in the study of polarization bias, given their widespread use filtering information spaces. Polarization is a social phenomenon, with serious consequences, real-life, particularly on media. Thus it to understand how machine learning algorithms, especially systems, behave polarized environments. In this paper, we within context users' interactions space items affects systems. We first formalize concept based item ratings then relate reviews investigate any potential correlation. propose domain independent data science pipeline automatically detect using rather than typical properties used polarization, such as item's content or network topology. perform an extensive comparison measures several benchmark sets show that our detection framework can different degrees outperforms existing capturing intuitive notion polarization. Our work essential step toward quantifying detecting ongoing sets, end, developed compute prevalence sets. It hope will contribute supporting future research emerging topic designing studying behavior