Scalable Fair Influence Maximization

Authors: Xiaobin Rui, Zhixiao Wang, Jiayu Zhao, Lichao Sun, Wei Chen

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We carry out detailed experimental analysis on five real social networks to study the trade-off between fairness and total influence spread. We test different fairness parameters, influence probabilities, seed budget, and community structures to confirm the performance of our proposed algorithms.
Researcher Affiliation Collaboration Xiaobin Rui China University of Mining and Technology Xuzhou, Jiangsu, China ruixiaobin@cumt.edu.cn Zhixiao Wang China University of Mining and Technology Xuzhou, Jiangsu, China zhxwang@cumt.edu.cn Jiayu Zhao China University of Mining and Technology Xuzhou, Jiangsu, China zhaojy@cumt.edu.cn Lichao Sun Lehigh University Bethlehem, PA, USA lis221@lehigh.edu Wei Chen Microsoft Research Asia Beijing, China, weic@microsoft.com
Pseudocode Yes Algorithm 1: RR-Generate: Generate RR sets Algorithm 2: FIMM: Fair Influence Maximization
Open Source Code No The paper does not contain any explicit statement or link indicating the availability of open-source code for the described methodology.
Open Datasets Yes Email The Email dataset [29] is generated using email data from a large European research institution... Flixster The Flixster dataset [30] is a network of American social movie discovery services... Amazon The Amazon dataset [31] is collected based on Customers Who Bought This Item Also Bought feature of the Amazon website... Youtube The Youtube dataset [31] is a network of the video-sharing web site... DBLP The DBLP dataset [31] is the co-authorship network...
Dataset Splits No The paper refers to using datasets but does not provide specific details on how they were split into training, validation, and test sets, or specify cross-validation settings.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. It only mentions the software used for implementation.
Software Dependencies Yes Note that our algorithm is currently implemented in Matlab 2022a, thus it costs more time to generate RR sets (generating RR sets in C++ could be at least 100 times faster).
Experiment Setup Yes In all tests, we run 10,000 Monte-Carlo simulations to evaluate both the influence spread and the fair influence under IC model. We also test influence probability p, inequality aversion parameter α and the seed budget k. ... For Email network, we set α = 0.5. ... We test different probabilities that range from 0.001 to 0.01 with the step of 0.001. ... For the Flixster network, we test the inequality aversion parameter α which ranges from 0.1 to 0.9 with the step of 0.1. We set k = 50 for both networks. ... For Amazon, Youtube, and DBLP networks, we set α = 0.5 and p(vi, vj) = 1/din(vj) ... We test different seed budget k that ranges from 5 to 50 with the step of 5.