Analysing the Evolution of Knowledge Graphs for the Purpose of Change Verification

Abstract

Knowledge graphs (KGs) are a core component of many web-based applications. KGs store information about entities such as persons and organizations. A central challenge is to keep KGs up-to-date while the entities in the real world continuously change. While the majority of the changes are correct, the KGs still receive erroneous changes due to vandalism and carelessness. Thus, change verification is required to ensure a quality of the information stored in the KG. Since manual change verification is labor intensive, different works have dealt with automatic change verification in the past. However, these works have not shed light on the evolutionary patterns of the KGs. Since the analysis of the evolution of social networks has contributed to link prediction between persons, we assume that the evolutionary patterns of KGs can contribute to the task of change verification. In this paper, we analyze the evolution of a KG, focusing on its topological features such as degree. The analysis reveals that the evolutionary patterns are similar to those of social networks. Subsequently, we develop classifiers that judge whether each incoming change is correct or incorrect. In the classifiers, we use a set of novel features, which originate from topological features of the KG. Finally, our experiments demonstrate that the novel features improve the verification performance. The results of this paper can contribute to making the KG editing process more efficient and reliable.