12:40 PM - 1:00 PM
[4L2-GS-4-03] Discriminative and Discrete Bayesian Gaussian Process Latent Variable Model to visualize user preference over latent item space
Keywords:Latent Variable Model, Recommendation, Bayesian Inference
Visualizing user preferences over an item space is essential for a recommendation system and the user.
The system can provide evidence of its recommendations, and the user can make informed decisions.
Dimension reduction techniques, such as GPLVM, t-SNE, and PCA, are commonly used to visualize items with high-dimensional features.
However, those models are unsupervised (not discriminative), so they do not encourage reflecting user rating labels to the low-dimensional space. In addition, although items have discrete features, those models assume continuous variables.
This study mainly addresses two issues: discriminability and discrete feature treatment.
We propose a Discriminative and Discrete Bayesian Gaussian Process Latent Variable Model (DDGPLVM). This model accepts discrete item features and user ratings and generates preference distributions over the continuous latent item space.
We confirmed our model using synthetic and real-world datasets. The findings demonstrate that our model generates discriminative and interpretable preference surfaces in the latent space.
The system can provide evidence of its recommendations, and the user can make informed decisions.
Dimension reduction techniques, such as GPLVM, t-SNE, and PCA, are commonly used to visualize items with high-dimensional features.
However, those models are unsupervised (not discriminative), so they do not encourage reflecting user rating labels to the low-dimensional space. In addition, although items have discrete features, those models assume continuous variables.
This study mainly addresses two issues: discriminability and discrete feature treatment.
We propose a Discriminative and Discrete Bayesian Gaussian Process Latent Variable Model (DDGPLVM). This model accepts discrete item features and user ratings and generates preference distributions over the continuous latent item space.
We confirmed our model using synthetic and real-world datasets. The findings demonstrate that our model generates discriminative and interpretable preference surfaces in the latent space.
Authentication for paper PDF access
A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.