Fairly Recommending with Social Attributes: A Flexible and Controllable Optimization Approach

Authors: Jinqiu Jin, Haoxuan Li, Fuli Feng, Sihao Ding, Peng Wu, Xiangnan He

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on two realworld datasets validate the effectiveness of our approach.
Researcher Affiliation Collaboration 1University of Science and Technology of China 2Peking University 3Beijing Technology and Business University
Pseudocode Yes Algorithm 1: Social-Aware Flexible Fair Recommendation with Controllable Accuracy (So FA)
Open Source Code Yes Our codes are available at https://github.com/mitao-cat/nips23_social_igf.
Open Datasets Yes Kuai Rec is a short-video recommendation dataset on the video-sharing platform Kuaishou3... Epinions is a dataset derived from a trust network...
Dataset Splits Yes Throughout our experiment, we randomly split the interactions of each dataset into training set (60%), validation set (20%), and testing set (20%).
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory amounts used for the experiments.
Software Dependencies No The paper mentions using 'BPRMF [9] as the backbone' but does not specify version numbers for programming languages, libraries, or other software dependencies required to reproduce the experiments.
Experiment Setup Yes For all compared methods, we use a pre-trained BPRMF [9] as the backbone with fixed optimal batch size and regularization coefficient, then fine-tune the model by using the baselines methods for achieving IGF notions... The learning rate for all methods is searched within {0.03lr, 0.1lr, 0.3lr, lr, 3lr, 10lr}, and the coefficients of IGF terms for regularization-based and post-processing baselines are searched within {0.1, 0.2, 0.5, 1, 2, 5}. The early stopping strategy is employed when the F1SP or F1EO value does not decrease over 5 epochs.