Fair Re-Ranking Recommendation Based on Debiased Multi-graph Representations.

Advanced Data Mining and Applications: 19th International Conference, ADMA 2023, Shenyang, China, August 21–23, 2023, Proceedings, Part I(2023)

Cited 0|Views12
No score
The successful application of graph neural networks in recommendation scenarios causes serious exposure of sensitive information of users. Research shows that social bias such as sexism and ageism are prevalent in recommendations, and the use of multi-graph information even makes it worse. Existing fair recommendation algorithms only concentrate on users’ sensitive attributes in user-item graph, failing to fully remove those attributes from multiple graphs. In addition, merely hiding sensitive information is not enough, there is still a gap in recommendation utility for different user groups.    In this work, we propose a novel fair re-ranking recommendation model based on debiased multi-graph representations, which contains three functional layers. Multi-graph embedding layer iteratively propagates and aggregates both topological and interactive information on multiple graphs. Attribute hiding layer uses generative adversarial networks to hide user sensitive information and thus debias users’ representations. Fair ranking layer adopts a re-ranking strategy with our proposed unfairness metric to further optimize the final recommendation list. Extensive experiments on real-world datasets demonstrate the performance of our proposed model in both recommendation utility and fairness, outperforming state-of-the-art models.
Translated text
Key words
AI Read Science
Must-Reading Tree
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined