Neural Networks for Personalized Recommender Systems

Download files
Access & Terms of Use
open access
Copyright: Zhang, Hangbin
Abstract
The recommender system is an essential tool for companies and users. A successful recommender system not only can help companies promote their products and services, but also benefit users by filtering out unwanted information. Thus, recommender systems are growing to be indispensable in a wide range of industries. Moreover, due to the fact that neural networks have been proved to be efficient and scalable, they are widely studied and applied to various fields. This thesis aims at developing methods for recommender systems by adapting neural networks. By exploring to adapt neural networks to recommender systems, this thesis investigates challenges that recommender systems are facing, and presents approaches to these challenges. Specifically, these challenges include: (1) data sparsity, (2) the complex relationships between users and items, (3) dynamic user preferences. To address the data sparsity, this thesis proposes to learn both collaborative features and content representations to generate recommendations in case of sparse data. Moreover, it proposes an architecture for the training process to further improve the quality of recommendations. To dynamically learn users' preferences, the thesis proposes to learn temporal features to capture dynamic changes of users' preferences. In this way, both the users' general preferences and the latest interactions are considered. To learn the complex relationships, this thesis also proposes a geometric method to measure nonlinear metric to learn the complex relationship among users and items. Moreover, the relationships between items are also considered to avoid potential problems.
Persistent link to this record
Link to Publisher Version
Additional Link
Supervisor(s)
Creator(s)
Editor(s)
Translator(s)
Curator(s)
Designer(s)
Arranger(s)
Composer(s)
Recordist(s)
Conference Proceedings Editor(s)
Other Contributor(s)
Corporate/Industry Contributor(s)
Publication Year
2021
Resource Type
Thesis
Degree Type
PhD Doctorate
UNSW Faculty
Files
download public version.pdf 2.47 MB Adobe Portable Document Format
Related dataset(s)