Ph.D.DS Dissertation Proposal | Di You | Wednesday, Nov. 8th @ 8:00am EST
8:00 a.m. to 9:30 a.m.
United States
DATA SCIENCE
Ph.D. Dissertation Proposal
Di You, Ph.D. Candidate
Wednesday, Nov. 8th, 2023 | 8:00AM - 9:30AM EST
Zoom Link: https://wpi.zoom.us/j/6735453923
Dissertation Committee:
Dr. Kyumin Lee, Professor, WPI. Advisor
Dr. Xiaozhong Liu, Professor, WPI
Dr. Randy Paffenroth, Professor, WPI
Yang Feng, Professor, Institute of Computing Technology, Chinese Academy of Sciences (ICT/CAS)
Title: Enhancing User Behavior-based Recommendations with Advanced Mathematics
Abstract:
Recommender Systems (RecSys) play a vital role in alleviating information overload, enhancing user experiences by filtering and pinpointing relevant content. They offer personalized suggestions towards candidate items tailored to meet user preferences in various application domains. The basic idea of recommender systems is to make use of the interactions between users and items and their associated side information (e.g., item titles or descriptions, user profiles, and user reviews for items), to predict the matching. However, conventional recommender systems face challenges such as data sparsity, bias and cold-start problem.
In this dissertation proposal, we aim to explore strategies from three perspectives to address these challenges and elevate users' satisfaction. Firstly, we proposed to fuse users' long-term interests modeling and short-term interests modeling into one model using a Quaternion-based gated mechanism, which benefits users' preference modeling in terms of both efficiency and effectiveness. Secondly, we make attempts with recent advances in graph neural networks (GNNs) and incorporate multi-behavior (e.g., click, page-view, cart) insights for alleviating the influence of data sparsity. Thirdly, we analysis the conventional recommendation from the perspective of causal effect and propose a novel causality-based architecture to compensate for the bias.