User Representation Learning for Personalized Recommendation Systems

  • Author / Creator
    Li, Chenglin
  • Recommendation systems have become an indispensable part of our lives, providing an effective solution to information overload and enhancing user satisfaction by suggesting the most relevant content. In recent years, deep learning has facilitated the creation of recommendation technologies that rely on sophisticated deep neural networks to learn intricate user representations, resulting in remarkable success. However, the existence of persistent issues, such as data sparsity and various biases, presents significant obstacles for deep neural methods in the pursuit of creating effective and unbiased recommendation systems.

    Cross-domain recommendation (CDR) is a promising approach that leverages data and knowledge from auxiliary domains to improve the performance of recommender systems when the data in the target domains is sparse. To address the challenge of cross-domain sequential recommendation with limited overlapping users, we developed a novel CDR method inspired by the real-world needs of industrial companies. Our method offers an effective solution to this problem. Furthermore, we extended our CDR approach to multi-target scenarios, where the objective is to enhance recommendation performance for three or more domains simultaneously, and we tackled the issue of negative transfer in this context.

    The utilization of a temporal heterogeneous graph is a compelling technique for capturing the intricate interactions between users and items in recommendation systems. In order to achieve precise sequential recommendations through the use of temporal graphs, we introduce a novel continuous-time representation learning model that can extract high-quality user and item representations from a temporal heterogeneous information network. Moreover, to address the issue of biases in recommendation systems, we propose an unbiased sequential recommendation model that incorporates the potential outcome framework (POF) and employs a disentangled graph transformer on a temporal user-item interaction graph to enhance performance.

  • Subjects / Keywords
  • Graduation date
    Fall 2023
  • Type of Item
  • Degree
    Doctor of Philosophy
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.