Profiling Users and Knowledge Graphs on the Web
Profiling refers to the process of collecting useful information or patterns about something. Due to the growth of the web, profiling methods play an important role in different applications such as recommender systems. In this thesis, we first demonstrate how knowledge graphs (KGs) enhance profiling methods. KGs are databases for entities and their relations. Since KGs have been developed with the objective of information discovery, we assume that they can assist profiling methods. To this end, we develop a novel profiling method using KGs called Hierarchical Concept Frequency-Inverse Document Frequency (HCF-IDF), which combines the strength of traditional term weighting method and semantics in a KG. HCF-IDF represents documents as a set of entities and their weights. We apply HCF-IDF to two applications that recommends researchers and scientific publications. Both applications show HCF-IDF captures topics of documents. As key result, the method can make competitive recommendations based on only the titles of scientific publications, because it reveals relevant entities using the structure of KGs. While the KGs assist profiling methods, we present how profiling methods can improve the KGs. We show two methods that enhance the integrity of KGs. The first method is a crawling strategy that keeps local copies of KGs up-to-date. We profile the dynamics of KGs using a linear regression model. The experiment shows that our novel crawling strategy based on the linear regression model performs better than the state of the art. The second method is a change verification method for KGs. The method classifies each incoming change into a correct or incorrect one to mitigate administrators who check the validity of a change. We profile how topological features influence on the dynamics of a KG. The experiment demonstrates that the novel method using the topological features can improve change verification. Therefore, profiling the dynamics contribute to the integrity of KGs.