Source-linked AI summary
Effect of initial configuration on network-based recommendation
Tao Zhou, Luo-Luo Jiang, Ri-Qi Su, Yi-Cheng Zhang
TL;DR
The paper addresses recommendation under information overload by proposing a network-based algorithm whose performance depends on how initial resource is configured. It introduces degree-dependent initialization and finds that reducing resource assigned to popular objects improves accuracy and personalization relative to uniform initialization.
Problem
Recommendation systems must filter information overload while uncovering users’ individual preferences, but accuracy and personalization need to be evaluated together.
Method
The authors construct a weighted object network, propagate user-specific initial resource through it, and regulate that initialization with a degree-dependent parameter β.
Results
The network-based method achieves ranking scores of 0.107, 0.122, and 0.140 for network-based recommendation, collaborative filtering, and global ranking, respectively; optimizing β further reduces the ranking score by 9%.
Takeaways & Limitations
Using an appropriate negative β outperforms uniform initialization by producing recommendations that are more accurate, less popular, and more personalized.
Abstract
from arXiv · showhide
In this paper, based on a weighted object network, we propose a recommendation algorithm, which is sensitive to the configuration of initial resource distribution. Even under the simplest case with binary resource, the current algorithm has remarkably higher accuracy than the widely applied global ranking method and collaborative filtering. Furthermore, we introduce a free parameter $β$ to regulate the initial configuration of resource. The numerical results indicate that decreasing the initial resource located on popular objects can further improve the algorithmic accuracy. More significantly, we argue that a better algorithm should simultaneously have higher accuracy and be more personal. According to a newly proposed measure about the degree of personalization, we demonstrate that a degree-dependent initial configuration can outperform the uniform case for both accuracy and personalization strength.