Adversarial and Contrastive Variational Autoencoder for Sequential Recommendation

By Zhe Xie, Chengxuan Liu, Yichi Zhang, Hongtao Lu, Dong Wang, Yue Ding. 2021

124 citations recorded by the source corpus.

Sequential recommendation as an emerging topic has attracted increasing attention due to its important practical significance. Models based on deep learning and attention mechanism have achieved good performance in sequential recommendation. Recently, the generative models based on Variational Autoencoder (VAE) have shown the unique advantage in collaborative filtering. In particular, the sequential VAE model as a recurrent version of VAE can effectively capture temporal dependencies among items in user sequence and perform sequential recommendation. However, VAE-based models suffer from a common limitation that the representational ability of the obtained approximate posterior distribution is limited, resulting in lower quality of generated samples. This is especially true for generating sequences. To solve the above problem, in this work, we propose a novel method called Adversarial and Contrastive Variational Autoencoder (ACVAE) for sequential recommendation. Specifically, we first introduce the adversarial training for sequence generation under the Adversarial Variational Bayes (AVB) framework, which enables our model to generate high-quality latent variables. Then, we employ the contrastive loss. The latent variables will be able to learn more personalized and salient characteristics by minimizing the contrastive loss. Besides, when encoding the sequence, we apply a recurrent and convolutional structure to capture global and local relationships in the sequence. Finally, we conduct extensive experiments on four real-world datasets. The experimental results show that our proposed ACVAE model outperforms other state-of-the-art methods.

Subfields: Computer Vision and Pattern Recognition, Information Systems, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Tat-seng Chua
2. Stefano Ermon
3. Jiaming Song
4. Xiangnan He
5. Chenlin Meng
6. Luc Van Gool
7. Dacheng Tao
8. Ming-Hsuan Yang

In Information Systems

1. Xiangnan He
2. Tat-seng Chua
3. Fuli Feng
4. Hongzhi Yin
5. Tong Chen
6. Xin Xia
7. Junliang Yu
8. Xiang Wang

In Representation & Self-Supervised Learning

1. Xiangnan He
2. Stefano Ermon
3. Jiaming Song
4. Fuli Feng
5. Tat-seng Chua
6. Abhishek Sinha
7. Chenlin Meng
8. Zi-Liang Huang

Citations in the corpus

• D2C: Diffusion-Decoding Models for Few-Shot Conditional Generation (2021, methodology): “In order to train a D2C model, we use an inference model qφ(z((1))|x) that predicts proper z((1)) latent variables from x; this can directly incorporate SSL methods [99], leading to the following objective: LD2C(θ, φ;w) := LD2(θ, φ;w) + λLC(qφ), (4) LD2(θ, φ;w) := Ex∼pdata,z(1)∼qφ(z(1)|x)[− log pθ(x”
• Self-Supervised Learning for Recommender Systems: A Survey (2022, background): “Feature Shuffling [29], [30], [76] switches rows and columns in the feature matrix X .”
• Diffusion Recommender Model (2023, methodology): “From Table 7, we have the following observations: 1) T-DiffRec and LT-DiffRec perform better than DiffRec and L-DiffRec by a large margin, justifying the effectiveness of the proposed timeaware reweighting strategy on temporal modeling; 2) the superior performance of T-DiffRec and LT-DiffRec than AC”