Aligning Dual Disentangled User Representations from Ratings and Textual Content

Tran, Nhu-Thuat; Lauw, Hady W.

doi:10.1145/3534678.3539474

Cited by 8 publications

(4 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Matrix Factorization VBPR [15], GraphCAR [62], VMCF [38], AMR [47], ACF [6], ConvMF [20], DeepCoNN [70], DVBPR [19] Multilayer Perceptron JRL [69] (Variational) Autoencoder CKE [66], ADDVAE [50], MVGAE [64] Attention Network UVCAN [30], ACF [6], MAML [27], MMRec [61], DMRL [26], IMRec [63] Graph Neural Network(GNN) HHFAN [4], MKGAT [46], MMGCN [59], DualGNN [53], MGAT [49], MEGCF [28], GRCN [58], Lattice [67], FREEDOM [73], HUIGN [57],…”

Section: Methods Modelsmentioning

confidence: 99%

“…Classical methods express the user preference as a single vector in the latent space which would not express the meaning facets. ADDVAE [50] utilizes the disentangled representations to capture better user preference that is influenced by several hidden factors. it leverages the text content to learn the second disentangled representation, coupling disentangled factors from two MacridVAE networks through mutual information maximization and then using the attention mechanism to align the representations with each other, which could resolve the sparsity of user-item interactions and map the uninterpretable dimensions from representations to words.…”

Section: Variational Autoencodermentioning

confidence: 99%

“…The framework is implemented by Pytorch and the input format is simple with textural embedding, visual embedding and the interaction graph. The following models except ADDVAE [50] are already implemented in our framework and the experiments are conducted to compare their recommendation performance. note that BPR and LightGCN are not using multimodal information but only use historical user-item interactions.…”

Section: Baseline Modelsmentioning

confidence: 99%

“…• ADDVAE [50] utilizes the MacridVAE network with mutual information maximization to get two disentangled representations from user-item interactions and the text content of users.…”

Section: Baseline Modelsmentioning

confidence: 99%

See 3 more Smart Citations

A Comprehensive Survey on Multimodal Recommender Systems: Taxonomy, Evaluation, and Future Directions

Zhou¹,

Zhou²,

Zeng³

et al. 2023

Preprint

View full text Add to dashboard Cite

Recommendation systems have become popular and effective tools to help users discover their interesting items by modeling the user preference and item property based on implicit interactions (e.g., purchasing and clicking). Humans perceive the world by processing the modality signals (e.g., audio, text and image), which inspired researchers to build a recommender system that can understand and interpret data from different modalities. Those models could capture the hidden relations between different modalities and possibly recover the complementary information which can not be captured by a uni-modal approach and implicit interactions. The goal of this survey is to provide a comprehensive review of the recent research efforts on the multimodal recommendation. Specifically, it shows a clear pipeline with commonly used techniques in each step and classifies the models by the methods used. Additionally, a code framework has been designed that helps researchers new in this area to understand the principles and techniques, and easily runs the SOTA models. Our framework is located at: https://github.com/enoche/MMRec.

show abstract

Section: Methods Modelsmentioning

confidence: 99%

Section: Variational Autoencodermentioning

confidence: 99%

Section: Baseline Modelsmentioning

confidence: 99%

“…• ADDVAE [50] utilizes the MacridVAE network with mutual information maximization to get two disentangled representations from user-item interactions and the text content of users.…”

Section: Baseline Modelsmentioning

confidence: 99%

See 2 more Smart Citations