Since its publication, R2RML has provided us with a powerful tool for generating RDF from relational data, not necessarily manifested as relational databases. R2RML has its limitations, which are being recognized by W3C's Knowledge Graph Construction Community Group. That same group is currently developing a specification that supersedes R2RML in terms of its functionalities and the types of resources it can transform into RDF--primarily hierarchical documents. The community has a good understanding of problems of relational data and documents, even if they might need to be approached differently because of their different formalisms. In this paper, we present a challenge that has not been addressed yet for relational databases--generating literals based on (outer-)joins. We propose a simple extension of the R2RML vocabulary and extend the reference algorithm to support the generation of literals based on (outer-)joins. Furthermore, we implemented a proof-of-concept and demonstrated it using a dataset built for benchmarking joins. While it is not (yet) an extension of RML, this contribution informs us how to include such support and how it allows us to create self-contained mappings rather than relying on less elegant solutions.

Supporting Relational Database Joins for Generating Literals in R2RML

Nowadays, Knowledge Graphs are extensively created using very different techniques, mapping languages among them. The wide variety of use cases, data peculiarities, and potential uses has had a substantial impact in how these languages have been created, extended, and applied. This situation is closely related to the global adoption of these languages and their associated tools. The large number of languages, compliant tools, and usually the lack of information of the combination of both leads users to use other techniques to construct Knowledge Graphs. Often, users choose to create their own ad hoc programming scripts that suit their needs. This choice is normally less reproducible and maintainable, what ultimately affects the quality of the generated RDF data, particularly in long-term scenarios. We devise with mapping translation an enhancement to the interoperability of existing mapping languages. This position paper analyses the possible language translation approaches, presents the scenarios in which it is being applied and discusses how it can be implemented.

Devising Mapping Interoperability with Mapping Translation

Nowadays, Knowledge Graphs (KG) are among the most powerful mechanisms to represent knowledge and integrate data from multiple domains. However, most of the available data sources are still described in heterogeneous data structures, schemes, and formats. The conversion of these sources into the desirable KG requires manual and time-consuming tasks, such as programming translation scripts, defining declarative mapping rules, etc. In this vision paper, we analyze the trends regarding the automation of KG construction but also the use of declarative mapping languages for the same process, and align the two by analyzing their tasks and a few exemplary tools. Our aim is not to have a complete study but to investigate if there is potential in this direction and, if so, to discuss what challenges we need to address to guarantee the maintainability, explainability, and reproducibility of the KG construction.

Declarative Description of Knowledge Graphs Construction Automation: Status & Challenges

Social media is one of the valuable information sources which present much data to the researchers. This information is mainly analyzed by machine learning and the deep learning methods, which lack semantics and interpretation in their outputs. Also, much attention is paid to the feature engineering there. We present a taxonomy of the different feature categories. The categories relate to the features learned during the training for analyzing the textual information, specifically available on social platforms. The ontological view of the data will represent knowledge in a more understandable form besides interpreting the machine learning results for various tasks related to the social data analysis. We chose Depression as the use case purpose. The ontology is designed using Ontology Web Language and Resource Description Framework in the Protégé. The validation of the ontology is carried out with designed competency questions.

FeatureOnto: A Schema on Textual Features for Social Data Analysis

Knowledge Graph construction is typically a task within
larger workflows, with a tight coupling between the abstract workflow and its execution.
Mapping languages increase interoperability and reproducibility of the mapping process, however, this should be extended over the entire Knowledge Graph construction workflow.
In this paper, we introduce an interoperable and reproducible solution for defining Knowledge Graph construction workflows leveraging Semantic Web technologies. We describe how a data flow workflow can be described interoperable (i.e., independent from the underlying technology stack) and reproducible (i.e., with detailed provenance) by composing semantic abstract function descriptions; and how such a semantic workflow can be automatically executed across technology stacks. We demonstrate that composing functions using the Function Ontology allows for functional descriptions of entire workflows, automatically executable using a Function Ontology Handler implementation. The semantic descriptions allow for interoperable workflows, the alignment with P-PLAN and PROV-O allows for reproducibility, and the mapping to concrete implementations allows for automatic execution.

Implementation-independent Knowledge Graph Construction Workflows using FnO Composition

When evolving datasets are used to generate a knowledge graph, it is usually challenging to keep the graph synchronized in a timely manner when changes occur in the source data. Current approaches fully regenerate a knowledge graph in such cases, which may be time consuming depending on the data type, size, and update frequency. We propose a continuous knowledge graph generation approach that can be applied on different types of data sources. We describe continuously updating knowledge graph versions represented as a Linked Data Events Stream, and use an RML processor for RDF generation. In this paper, we present our approach and demonstrate it on different types of data such as bike-sharing, public transport timetables, and weather data. By describing entities with unique, immutable, and reproducible iris, we were able to identify changes in the original data collection, reducing the number of materialized triples and generation time. Our use-cases show the importance of mechanisms to derive unique and stable IRI strategies of data source updates, to enable efficient knowledge graph generation pipelines. In the future, we will extend our approach to handle deletions in data collections, and conduct an extensive performance evaluation.

Continuous generation of versioned collection's members with RML and LDES

Knowledge Graphs (KGs) play a significant and growing role for semantics-based support of a wide variety of applications. Until recently, creating and maintaining such knowledge graphs was done in a one-off manner requiring significant manual effort and expertise. Over the last few years, the first KG management platforms supporting the lifecycle of KGs from their creation to their maintenance and use have appeared. In this paper,  we first survey these platforms. We then take a step further and identify common functionalities across such platforms. We  discuss nineteen such functionalities categorized into four groups: creating, extending, using, and maintaining KGs. Based on the findings of this analysis, we present our proposed KG management platform for the biodiversity domain, iKNOW. We focus on the architecture and the KG creation workflow, but also touch on other aspects.

What is needed in a Knowledge Graph Management Platform? A survey and a proposal

Knowledge workers' personal and work related concepts (e.g. persons, projects, topics) are usually not sufficiently covered by knowledge graphs. Yet, already handmade classification schemes, prominently folder structures, naturally mention several of their concepts in file names. Thus, such data could be a promising source for constructing personal knowledge graphs. However, this idea poses several challenges: file names are usually noisy non-grammatical text snippets, while folder structures do not clearly define how concepts relate to each other. To cope with this semantic gap, we include knowledge workers as humans-in-the-loop to guide the building process with their feedback. Our semi-automatic personal knowledge graph construction approach consists of four major stages: domain term extraction, ontology population, taxonomic and non-taxonomic relation learning. We conduct a case study with four expert interviews from different domains in an industrial scenario. Results indicate that file systems are promising sources and, combined with our approach, already yield useful personal knowledge graphs with moderate effort spent.

A Human-in-the-Loop Approach for Personal Knowledge Graph Construction from File Names

The majority of available datasets in open government data are statistical. They are widely published by different governments to be used by the public and data consumers. However, most datasets in open data portals are not provided in RDF format. Moreover, the datasets are isolated from one another while conceptually connected. Through this paper, a knowledge graph is constructed for the disease-related datasets of a Canadian government data portal, Nova Scotia Open Data. We transformed all the disease-related datasets to RDF according to the Semantic Web standards and enriched them by semantic rules and an external ontology. The ontology designed to develop the graph adheres to best practices and standards, allowing for expansion, modification and flexible re-use (https://zenodo.org/record/5517236#.Ye_MsfXMJb8). The study also discusses the lessons learned during the cross-dimensional knowledge graph construction and integrating open statistical datasets from multiple sources. 


Constructing a Knowledge Graph from Open Statistical Data: The Case of Nova Scotia DiseaseDatasets

This paper presents the Coronavirus Disease Ontology (CovidO), a superset of the available Coronavirus (Covid-19) ontologies, including all the possible dimensions. CovidO consists of an ontological network of thriving distinct dimensions for storing coronavirus information. CovidO has 175 classes, 169 properties, 4141 triples, 645 individuals with 264 nodes and 308 edges. CovidO is based on standard input of coronavirus disease data sources, activities, and related sources, which collects and validates records for decision-making used to set guidelines and recommend resources. We present CovidO to a growing community of artificial intelligence project developers as pure metadata and illustrate its importance, quality, and impact. The ontology developed in this work addresses grouping the existing ontologies to build a global data model.

The Coronavirus Disease Ontology (CovidO)

Recent advances in neural networks have solved common graph problems such as link prediction, node classification, node clustering, node recommendation by developing embeddings of entities and relations into vector spaces.  Graph embeddings encode the structural information present in a graph. The encoded embeddings then can be used to predict the missing links in a graph. However, obtaining the optimal embeddings for a graph can be a computationally challenging task specially in an embedded system. Two techniques which we focus on in this work are 1) node embeddings from random walk based methods and 2) knowledge graph embeddings. Random walk based embeddings are computationally inexpensive to obtain but are sub-optimal whereas knowledge graph embeddings perform better but are computationally expensive. In this work, we investigate a transformation model which converts node embeddings obtained from random walk based methods to embeddings obtained from knowledge graph methods directly without an increase in the computational cost. Extensive experimentation shows that the proposed transformation model can be used for solving link prediction in real-time.

KGCW 2022 Workshop Submissions