TY - GEN
T1 - Exploring the Role of Local and Global Explanations in Recommender Systems
AU - Radensky, Marissa
AU - Downey, Doug
AU - Lo, Kyle
AU - Popovic, Zoran
AU - Weld, Daniel S.
N1 - Funding Information:
This work is partially supported by The National Science&Technology Pillar Program (2008BAI50B03), National Natural Science Foundation of China (NO.10574140, 10925419, 90920302,10874203, 60875014).
Publisher Copyright:
© 2022 Owner/Author.
PY - 2022/4/27
Y1 - 2022/4/27
N2 - Explanations are well-known to improve recommender systems' transparency. These explanations may be local, explaining individual recommendations, or global, explaining the recommender model overall. Despite their widespread use, there has been little investigation into the relative benefits of the two explanation approaches. We conducted a 30-participant exploratory study and a 30-participant controlled user study with a research-paper recommender to analyze how providing local, global, or both explanations influences user understanding of system behavior. Our results provide evidence suggesting that both are more helpful than either alone for explaining how to improve recommendations, yet both appeared less helpful than global alone for efficiently identifying false positive and negative recommendations. However, we note that the two explanation approaches may be better compared in a higher-stakes or more opaque domain.
AB - Explanations are well-known to improve recommender systems' transparency. These explanations may be local, explaining individual recommendations, or global, explaining the recommender model overall. Despite their widespread use, there has been little investigation into the relative benefits of the two explanation approaches. We conducted a 30-participant exploratory study and a 30-participant controlled user study with a research-paper recommender to analyze how providing local, global, or both explanations influences user understanding of system behavior. Our results provide evidence suggesting that both are more helpful than either alone for explaining how to improve recommendations, yet both appeared less helpful than global alone for efficiently identifying false positive and negative recommendations. However, we note that the two explanation approaches may be better compared in a higher-stakes or more opaque domain.
KW - explainable AI
KW - human-AI interaction
UR - http://www.scopus.com/inward/record.url?scp=85129767101&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85129767101&partnerID=8YFLogxK
U2 - 10.1145/3491101.3519795
DO - 10.1145/3491101.3519795
M3 - Conference contribution
AN - SCOPUS:85129767101
T3 - Conference on Human Factors in Computing Systems - Proceedings
BT - CHI 2022 - Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems
PB - Association for Computing Machinery
T2 - 2022 CHI Conference on Human Factors in Computing Systems, CHI EA 2022
Y2 - 30 April 2022 through 5 May 2022
ER -