TY - GEN
T1 - Geo-Personalization Bias in News Search
T2 - 19th ACM International Conference on Web Search and Data Mining, WSDM 2026
AU - You, Jaebeom
AU - Hong, Seung Kyu
AU - Liu, Ling
AU - Lee, Kisung
AU - Kwon, Hyuk Yoon
N1 - Publisher Copyright:
© 2026 Owner/Author.
PY - 2026/2/21
Y1 - 2026/2/21
N2 - This study systematically examines the formation patterns of filter bubbles in search engine results from an end-user perspective, analyzing them across diverse perspectives. Specifically, we investigate filter bubble formation patterns in search results obtained from two globally dominant search engines over a period of two months on six representative controversial topics. We analyze filter bubble formation in search engine results by varying geo-location while controlling other influencing factors. To ensure objective and scalable analysis, we utilize annotations generated by large language models (LLMs) that represent multiple distinct perspectives. We provide evidence demonstrating how filter bubbles are formed by addressing the following four research questions (RQs): search engine-specific personalization algorithms (RQ1) differently contribute to filter bubble formation; 2) user geo-location differences (RQ2) generate distinct filter bubbles reflecting regional characteristics; 3) these filter bubbles evolve in response to regional social events over time (RQ3); and 4) topic characteristics (RQ4) result in varying patterns of filter bubble formation. Our source code and scripts are publicly available at: https://anonymous.4open.science/r/Filter_Bubble-183C.
AB - This study systematically examines the formation patterns of filter bubbles in search engine results from an end-user perspective, analyzing them across diverse perspectives. Specifically, we investigate filter bubble formation patterns in search results obtained from two globally dominant search engines over a period of two months on six representative controversial topics. We analyze filter bubble formation in search engine results by varying geo-location while controlling other influencing factors. To ensure objective and scalable analysis, we utilize annotations generated by large language models (LLMs) that represent multiple distinct perspectives. We provide evidence demonstrating how filter bubbles are formed by addressing the following four research questions (RQs): search engine-specific personalization algorithms (RQ1) differently contribute to filter bubble formation; 2) user geo-location differences (RQ2) generate distinct filter bubbles reflecting regional characteristics; 3) these filter bubbles evolve in response to regional social events over time (RQ3); and 4) topic characteristics (RQ4) result in varying patterns of filter bubble formation. Our source code and scripts are publicly available at: https://anonymous.4open.science/r/Filter_Bubble-183C.
KW - filter bubble detection
KW - geo-location personalization
KW - multi-perspective llm annotation
KW - multi-region data collection
KW - search engine personalization
KW - statistical significance testing
UR - https://www.scopus.com/pages/publications/105033148952
U2 - 10.1145/3773966.3777984
DO - 10.1145/3773966.3777984
M3 - Conference contribution
AN - SCOPUS:105033148952
T3 - WSDM 2026 - Proceedings of the 19th ACM International Conference on Web Search and Data Mining
SP - 850
EP - 859
BT - WSDM 2026 - Proceedings of the 19th ACM International Conference on Web Search and Data Mining
PB - Association for Computing Machinery, Inc
Y2 - 22 February 2026 through 26 February 2026
ER -