Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


22225.pdf (2.45 MB)

ETD Abstract Container

Abstract Header

Groupwise Distance Learning Algorithm for User Recommendation Systems

Zhang, Yi
http://orcid.org/0000-0003-4147-7131
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1471347509

Abstract Details

2016, PhD, University of Cincinnati, Engineering and Applied Science: Mechanical Engineering.
The dissertation study focuses on addressing some challenges in user-to-user recommendation area. As results, a new algorithm, Group-wise Distance Learning (GDL), is proposed in this study as a backbone to build user recommendation systems with an approximate personalization effect. By analyzing the established connections among users, the recommendation system empowered by GDL can induce users' social preference. Instead of learning individual user's social preference separately, GDL is designed to adopt an efficient strategy by clustering users into a number of groups. Users are grouped together in accordance with their similar social preference, which is extracted by GDL and encoded as feature-wise weights. Thanks for the approximate personalization with learned group-wise social preference, a GDL-empowered user recommendation system could help users more effectively expand their social networking. With the assistance of this new user recommendation system, a social network is expected to evolve faster and more connected in a collective manner. The performance of a user recommendation system empowered by GDL was compared with the recommendation system empowered by Nearest Neighbours (NN) in a series of simulation experiments. The simulation experiment was designed to mimic the evolution of a social network fueled by the interactions between users and the recommendation system. According to simulation experiment results based on two distinct synthetic data sets, the proposed GDL-empowered user recommendation system outperformed NN-empowered counterparts. Because of the cold-start issue, we recommend building a hybrid user recommendation system of these two methods. For a new user with zero user connections, use a NN-based method to generate recommendation initially. For users having a solid social profile, a GDL-empowered system can serve them better. In combination, a superior recommendation system can be built with benefits of two different methods.
Hongdao Huang, Ph.D. (Committee Chair)
Jing Shi, Ph.D. (Committee Member)
David Thompson, Ph.D. (Committee Member)
Xuefu Zhou, Ph.D. (Committee Member)
95 p.
Computer Science
recommendation system; user recommendation system; distance learning

Recommended Citations

Citations

  • Zhang, Y. (2016). Groupwise Distance Learning Algorithm for User Recommendation Systems [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1471347509

    APA Style (7th edition)

  • Zhang, Yi. Groupwise Distance Learning Algorithm for User Recommendation Systems. 2016. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1471347509.

    MLA Style (8th edition)

  • Zhang, Yi. "Groupwise Distance Learning Algorithm for User Recommendation Systems." Doctoral dissertation, University of Cincinnati, 2016. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1471347509

    Chicago Manual of Style (17th edition)

ucin1471347509
655
© 2016, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.