Relation extraction techniques for personal knowledge graphs
Ana B. Rios-Alvarado, Jose L. Martinez-Rodriguez · Institution of Engineering and Technology eBooks · 2023
Personal knowledge graphs (PKGs) refer to knowledge graphs (KGs) that contain information about personal descriptions. Thus, PKGs contain definitions and/or descriptions of information concerning a person or user. The development of PKGs is an essential aspect of decision-making applications where it may be necessary to involve user information such as preferences (tastes, styles, colors), physical characteristics (health, ergonomic), and historical (academic, legal), to mention a few. One of the main aspects of constructing such structures lies in the information components connected to form a graph. These components, called triples, are composed of Subject-Predicate-Object elements, where the Subject is oriented to be a real-world entity about a person, the Object is an entity or value that is said of the Subject, and the Predicate is the relationship that exists between these two things. However, the construction of PKGs is often a challenge, whether performed manually (by an expert) or automatically (computationally). On the one hand, their definition has not been widely studied, so the construction of PKGs is time-consuming even if performed manually. On the other hand, their computational construction is a process that, according to the source of information (and its structure), can implicate various language challenges to obtain the necessary triple elements accurately (to extract the relations that will be used in the construction and connection of triples). This chapter presents an overview of techniques for extracting relations applied in the construction of PKGs. The idea is to provide conceptual aspects on the extraction of relations for the construction of KGs and then to review the most common techniques applied to PKGs. Finally, some considerations in the extraction of relations and the construction of PKGs are presented.