Author: Shu, Honglin
Title: Leveraging graph neural network and meta learning for user cold-start recommendation
Degree: M.Sc.
Year: 2021
Subject: Recommender systems (Information filtering)
Neural networks (Computer science)
Hong Kong Polytechnic University -- Dissertations
Department: Department of Computing
Pages: ix, 53 pages : color illustrations
Language: English
Abstract: In the era of rapid generation of data and information, people can easily obtain information from Internet but at the same time they also encounter the problem of information overload. For solving part of the problem, recommendation systems have been widely used in e-commerce, social networks, streaming media, and other web services, such as Amazon, Twitter, Netflix, Pinterest, etc. Many companies rely on accurate personalized recommendations to improve user experience and boost companies' development. Personalized recommendations had already revolutionized various web services. Collaborative Filtering (CF) is perhaps the most popular and important approach in recommendation systems, and many novel methods are also improvements based on CF's idea. Although many CF-based methods are used in the recommendation system, they typically suffer from the challenge of cold-start recommendation when the historical interaction of new users is limited. To deal with the problem, many methods use implicit methods to generate user embedding and item embedding, and use various auxiliary information to alleviate the cold-start problem such as user profile, user's social relationship, item information, and even extracting information from the heterogeneous information network. In this thesis entitled "Leveraging Graph Neural Network and Meta Learning for User Cold-start Recommendation", a novel method to deal with user cold-start problem in recommendation systems is explored. Inspired by the distinctive properties of graph convolution network and meta learning, we propose a method that combines a graph auto-encoder-based model and an optimization-based meta learning method to alleviate the user cold-started problem. This method also equips with a random enclosing subgraph sampling, which can randomly sample k-hop enclosing subgraph as the meta learning tasks for existing users, and the r-decaying theory proves that low-hop enclosing subgraph can effectively approximate the heuristics from high-order subgraph. Besides, our experiments have proved the effectiveness of our method. Comparing with some benchmarking models on three data sets, the proposed model achieves competitive results without relying on the side information on two data sets.
Rights: All rights reserved
Access: restricted access

Files in This Item:
File Description SizeFormat 
5859.pdfFor All Users (off-campus access for PolyU Staff & Students only)1.28 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: