Access and use of semantically defined metadata based on RDF repositories can benefit several types of computational tasks. However, RDF triples tend to undergo modifications as new releases of the repositories appear, which implies a challenging scenario for RDF-based generated annotations. In this context, existing annotations need to be updated according to the evolution of undergoing knowledge base used for their definitions. In this paper, we propose an adaptation framework for updating semantic annotations defined from structured RDF data. Our adaptation approach relies on modifications detected in the evolution of RDF knowledge bases. We design and formalize adaptation operations which are applied to update annotation states. We present and formalize the framework and discuss existing open challenges in our research task.
The generation of metadata (data about data) on Web documents, videos and images
using existing Resource Description Framework (RDF) knowledge bases plays a key
role to computational systems. This type of metadata is called semantic annotations [
In the last years, a large number of interconnected RDF knowledge bases have
emerged describing various types of resources in a structured way, e.g., Dbpedia [
Existing literature has presented methodologies to address this problem. The
studies that address the automatic detection of inconsistent annotations [
In this article, we propose a framework capable of identifying and applying maintenance actions in semantic annotations affected by the evolution of RDF knowledge bases as automatically as possible. Our maintenance process comprises the execution of three steps (cf. Secion 2).
The remaining of this article is organized as follows: Section 2 presents our framework including its description and formalization; Section 3 provides a discussion on existing open challenges in our research. Finally, Section 4 draws the conclusion remarks. 2
The key contribution of this research consists of a framework capable of executing
modifications to update annotations as automatically as possible. We propose the
framework ANNOLOD to support maintaining instance annotations in RDF repositories. In
our study, we adapted the annotation model proposed by Cardoso et al. [
The adaptation was carried out, because the model defined in Cardoso et al. [
the need to add the attributes i, t, d, Of f set, rel, and p in the definition of an annotation
a. In the original model of Cardoso et al. [
The proposed framework defines annotation adaptation actions to be performed automatically when an RDF dataset used to create semantic annotations evolves (i.e., a new release is generated). These actions are necessary to keep annotations consistent and up to date over time. The necessary input consists of the interconnected initial datasets, being Rj and Rj+1 (its new release that can affect existing annotations) and the existing in place Aj annotations. The ultimate goal of the framework is to obtain the updated Aj+1 annotations according to the new release Rj+1 dataset. The framework performs a series of steps (cf. Algorithm 1). Each step is explained in further details (cf. Steps A, B, and C).
Require: R , Rj+1, Aj
j 1: Aj+1 ? 2: detectChanges(Rj ; Rj+1) recAf f Annotations( ; Aj ) recU nAf f Annotations( ; Aj ) j+1 unaff 3: Aaff 4: Aunaff 5: Aj+1 A [ A
7: Aj+1 Aj+1 [ applyAction( ; ai) 8: end for
Step A: this step consists of detecting a series of modifications that occurred in a given time period based on two releases of an RDF dataset (line 2 in Algorithm 1). This operation is known as because it computes the difference between the two datasets, recognizing added, removed, and not updated elements. Changes can be of the simple type (such as unit actions of adding or removing triples), or complex operations (update actions) of the knowledge stored in the datasets.
Step B: this step consists of recognizing and filtering the annotations affected by the changes of those that are not affected (lines 3 to 4 in Algorithm 1). An annotation
can be created, removed or updated. In our solution, annotations that share a subject (s) with a tk triple in which tk 2 are considered affected by change modifications. These are considered outdated annotations and maintenance actions on the annotations must be applied to them. We assume that unaffected annotations can be directly reused in composing the final set of updated annotations and added to the final set of Aj+1 annotations (line 5 in Algorithm 1). However, annotations classified as affected is further handled by our framework. This involves investigating which and how computed change operations influence the definition of existing annotations.
Step C: this step involves applying corrective actions (lines 6 to 8 in Algorithm 1) to the affected and outdated annotations detected in step B. For example, an action type may be a “reanotation”. In this case, an annotation ai 2 Aj defined on the basis of a triple tk adapted its subject (s). The framework generates as a final result a stable and semantically consistent set of annotations Aj+1 (line 9 in Algorithm 1) concerning the updated RDF data in the new dataset Rj+1. 3
The key research challenge at step A (cf. Section 2) is to understand what kind of changes at the level of instances can affect existing annotations. For example, to what extent does the removal of a triple RDF (used in defining an annotation) impact the consistency of such an annotation?
In step B (cf. Section 2), a key challenging research refers to how to accurately categorize an annotation as inconsistent based on computed change operations. In this sense, we need to investigate to which extent semantics defined in the annotations are affected by the observed changes. For example, the removal of the associated triple may be a typical case that affects the annotation. However, if there are other types of changes related to the triple subject, it is necessary to further understand how they make the annotation semantically inconsistent.
The main challenge in step C (cf. Section 2) refers to the definition and correct application of annotation adaptation actions to ensure the updating of semantically consistent affected annotations. The definition of actions (adaptation operations) requires investigating techniques that express the necessary conditions for their application. 4
The real value of semantically-enabled computer systems lays on the reliability of semantic annotations. This work studied how to keep RDF-based annotations up-to-date according to the evolution of RDF repositories. We proposed a framework for the (semi)automatic maintenance of semantic annotations affected by RDF data evolution. Our defined adaptation algorithm works on the basis of change operations automatically identified in the evolution of RDF datasets. We are currently further investigating the adaptation actions, their formalization and applicability. Next steps involve the full implementation of a software tool for the adaptation of RDF-based semantic annotations maintenance. We also plan to conduct thorough experimental analyses with real-world datasets. This work is supported by the São Paulo Research Foundation (FAPESP) (Grants #2017/023255, #2019/14582-8 and #2013/08293-7)3. 3 The opinions expressed in this work do not necessarily reflect those of the funding agencies.