Infer-AVAE: An attribute inference model based on adversarial variational autoencoder
Infer-AVAE: An attribute inference model based on adversarial variational autoencoder
复制标题
Infer-AVAE：基于对抗性变分自动编码器的属性推理模型
DOI：
10.1016/j.neucom.2022.02.006
复制
发表时间：
2022
期刊：
影响因子：
6
通讯作者：
Xiaohong Guan
中科院分区：
文献类型：
--
作者：
Yadong Zhou;Zhihao Ding;Xiaoming Liu;Chao Shen;Lingling Tong;Xiaohong Guan
用户属性，如性别和教育，在社交网络中面临严重的不完整性。属性推理的目的是根据观察到的数据推断用户丢失的属性标签，使这些有价值的数据可用于下游任务，如用户分析和个性化推荐。最近，变分自动编码器（VAE），一种端到端的深度生成模型，通过以半监督的方式处理问题，表现出了很好的性能。然而，VAE在应用于属性推断时容易遭受过拟合和过平滑。具体来说，VAE实现的多层感知器（MLP）只能重建输入数据，但无法推断丢失的部分。而使用趋势图神经网络（GNNs）作为编码器存在的问题是，GNNs聚集来自邻域的冗余信息并生成不可区分的用户表示，称为过平滑。在本文中，我们提出了一个属性推理模型的基础上的对抗性VAE（Infer-AVAE），以应对这些问题。具体来说，为了克服过度平滑，Infer-AVAE在编码器中统一了MLP和GNN，以分别学习正和负潜在表示。同时，训练对抗网络来区分这两种表示，并训练GNN通过对抗训练来聚合更少的噪声以获得更鲁棒的表示。最后，为了减轻过拟合，互信息约束被引入作为正则化器，以使解码器更好地利用表示中的辅助信息，并生成不受观测限制的输出。我们在四个真实的社交网络数据集上评估了我们的模型，实验结果表明，我们的模型在准确性上平均优于基线7.0%。
User attributes, such as gender and education, face severe incompleteness in social networks. Attribute inference aims to infer users’ missing attribute labels based on observed data to make this valuable data usable for downstream tasks like user profiling and personalized recommendation. Recently, variational autoencoder (VAE), an end-to-end deep generative model, has shown promising performance by handling the problem in a semi-supervised way. However, VAEs can easily suffer from over-fitting and over-smoothing when applied to attribute inference. Specifically, VAE implemented with multi-layer perceptron (MLP) can only reconstruct input data but fail to infer missing parts. While using the trending graph neural networks (GNNs) as encoder has the problem that GNNs aggregate redundant information from the neighborhood and generate indistinguishable user representations, known as over-smoothing. In this paper, we propose an attribute Inference model based on Adversarial VAE (Infer-AVAE) to cope with these issues. Specifically, to overcome over-smoothing, Infer-AVAE unifies MLP and GNNs in the encoder to learn positive and negative latent representations respectively. Meanwhile, an adversarial network is trained to distinguish the two representations, and GNNs are trained to aggregate less noise for more robust representations through adversarial training. Finally, to relieve over-fitting, mutual information constraint is introduced as a regularizer for the decoder to make better use of auxiliary information in representations and generate outputs not limited by observations. We evaluate our model on four real-world social network datasets, and experimental results demonstrate that our model averagely outperforms baselines by 7.0% in accuracy.