Collaborative Filtering Model via Contractive Auto-encoders

This presentation delves into collaborative filtering (CF) models, a crucial aspect of recommender systems, focusing on a hybrid approach that leverages contractive auto-encoders. It begins with an introduction to CF, emphasizing its role in personalized services and addressing the data sparsity challenge through hybrid CF models. The presentation highlights a specific study by Shuai, Lina, and Xiwei, who propose a hybrid model using contractive auto-encoders within a matrix factorization framework. It covers the application of Restricted Boltzmann Machines (RBMs) and the model's ability to capture non-linear properties from item content data. The presentation then contrasts conventional CF models with the hybrid approach, discussing the cold-start problem and the use of side information. Furthermore, it introduces aSDEA (another variant of SDAE), a deep learning model that integrates deep representation learning and matrix factorization. The presentation concludes with a discussion of future work, including potential enhancements with recurrent and convolutional neural networks, along with two effective hybrid CF models: AutoSVD++ and AutoSVD. Finally, the presentation provides a comprehensive bibliography of sources used.

Collaborative Filtering Model
via Contractive Auto-encoders

2
Introduction:
The following presentation demonstrates Collaborative
Filtering or CF. It is helpful to provide users with various personalized
services and products.

3
• Further, this leverages the side information of items and users that has
been extensively researched here to develop performance.

4
• To deal with the rising sparseness of the user-item matrix, it has been
still remaining in challenge.

5
• In order to tackle such issue, the hybrid kind of CF like assimilation
with filtering that is content based is used.

6
• Further with the rise in number of online information the
recommender system has been playing with more indispensable role.

7
• Moreover, it has been boosting the sales for organizations under e-
commerce.

8
• The user preference of CF is studied that has been extending from
usage patterns like user-item interactions for making
recommendations.

9
• Thus the hybrid methods have been getting more attention to tackle
those issues through assimilation the CF based and content based
models

10
• The study of Shuai, Lina and Xiwei a latest hybrid model through
creating contractive auto-encoder paradigm within the matrix
factorization framework is demonstrated.

11
• The wide range of experiments done in the study over three of the
large-scale real kind of datasets has been denoting the suggested
approach.

12
• Apart from this maximum of the approaches has been depending on
handcrafted developed features of engineering.