Hybrid deep-learning approach for document-context aware recommendation system
Traditional Collaborative Filtering (CF) methods such as Matrix Factorization (MF) only considers user preferences as a linear combination of user and item latent factors, leading to limited learning capabilities. To overcome this, many researchers have incorporated Deep Learning (DL) techniques int...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2021
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Traditional Collaborative Filtering (CF) methods such as Matrix Factorization (MF) only considers user preferences as a linear combination of user and item latent factors, leading to limited learning capabilities. To overcome this, many researchers have incorporated Deep Learning (DL) techniques into traditional CF. However, DLbased CF methods still suffer from data sparsity and cold-start problems. Several researchers have suggested incorporating side information into CF, which eventually converts the CF method to a Hybrid-based (HB) method. Besides, many HB methods also lack document based approaches, which have proven to improve recommendation accuracy by incorporating text-based document data. This thesis proposes a document context-aware Recommender System (RS) using a hybrid DL approach, namely Convolutional Neural Matrix Factorization ++ (CNMF++). CNMF++ is a hybrid DL model that jointly integrates three models including Generalized Matrix Factorization ++ (GMF++), Multilayer Perceptron ++ (MLP++) and Convolutional Neural Network (CNN). GMF++ and MLP++ first perform user-item metadata feature extraction by using Stacked Denoising Autoencoder (SDAE). |
|---|