Publication - Paper

Time-Difference-Labeled Path

Date

    12/2014 - 09/2015

Authors

    Manling Li, Yantao Jia, Yuanzhuo Wang, Zeya Zhao, Xueqi Cheng.

Sponsored by

    National Natural Science Foundation of China (NSFC) No. 61402442

Paper

Problem

    Predicting links and their building time in a knowledge network, i.e., a network with multiple typed vertices and time-labeled edges, is important to detect the evolution of a dynamic network and has been paid much attention. Actually, we may be more interested in “Will two authors co-write a paper within 5 years?” than “Will two authors co-write a paper?”. Nevertheless, the traditional structure-based methods, e.g., the Path Ranking Algorithm (PRA) (Lao et al., 2012), used the paths without the time information to predict the existence of links rather than the building time of links.

My Work

  • Came up with the idea to employ building time in a knowledge graph to detect the evolution of a dynamic graph, which can also improve the performance of link prediction.
  • Defined Time-Difference-Labeled Path, which is a time-involving path to integrate time information into relation paths.
  • Proposed a Time-Difference-Labeled Path based method (TDLP) to predict links and their building time. It regards the time information and the path information as a unified feature and models their interplay in an intrinsic way. Thus the building time of links can be learned.
  • The paper was published as a poster by AAAI 2016 (Lead Author).