Options
From Walls to Windows: Creating Transparency to Understand Filter Bubbles in Social Media
Type
conference contribution
Date Issued
2024-10-18
Author(s)
Research Team
Interactions Research Group (https://interactions.ics.unisg.ch)
Abstract
Social media platforms play a significant role in shaping public opinion and societal norms. Understanding this influence requires examining the diversity of content that users are exposed to. However, studying filter bubbles in social media recommender systems has proven challenging, despite extensive research in this area. In this work, we introduce SOAP (System for Observing and Analyzing Posts), a novel system designed to collect and analyze very large online platforms (VLOPs) data to study filter bubbles at scale. Our methodology aligns with established definitions and frameworks, allowing us to comprehensively explore and log filter bubbles data. From an input prompt referring to a topic, our system is capable of creating and navigating filter bubbles using a multimodal LLM. We demonstrate SOAP by creating three distinct filter bubbles in the feed of social media users, revealing a significant decline in topic diversity as fast as in 60min of scrolling. Furthermore, we validate the LLM analysis of posts through an inter-and intra-reliability testing. Finally, we open source SOAP as a robust tool for facilitating further empirical studies on filter bubbles in social media.
Language
English
Keywords
Filter Bubbles
Social Media
Black-Box testing
Systemic Risks
Deductive Coding
VLOP
DSA
HSG Classification
contribution to scientific community
Pages
12
Event Title
NORMalize 2024: The Second Workshop on the Normative Design and Evaluation of Recommender Systems, co-located with the ACM Conference on Recommender Systems 2024 (RecSys 2024)
Event Location
Bari, Italy
Event Date
October 18, 2024
Subject(s)
Division(s)
Contact Email Address
lukajurelars.bekavac@student.unisg.ch
File(s)
Loading...
open access
Name
Bekavac et al - 2024 - From Walls to Windows.pdf
Size
231.48 KB
Format
Adobe PDF
Checksum (MD5)
4121cacf2eb7d94a634c8ebe5d37a11e