Exponentially increasing inter-related scholarly knowledge is being published on multiple scholarly communication infrastructures. Retrieving data from a single scholarly communication infrastructure is not suficient to meet users complex requirements. Moreover, the manual linking of scholarly knowledge to produce inter-related outputs is a cumbersome task. Required are flexible and user-friendly mechanisms that retrieve inter-related data from distributed scholarly infrastructures. In the proposal presented here, we leverage a federated interface to access data from multiple scholarly communication infrastructures to answer complex user queries. Specifically, we use ORKG (Open Research Knowledge Graph), ORKG Ask, DataCite, OpenAIRE Graph and Semantic Scholar endpoints to access data from these infrastructures in a federated manner. We present the work for the information needs of diverse stakeholders to demonstrate the practicability of the federation, the straightforward implementation and the added value. The code of our service is publicly available on Gitlab1,2.
structured and semantic manner and enables retrieval of related scholarly content or derivatives thereof (e.g., comparisons). We propose that the (machine-readable) scholarly knowledge published in ORKG and other scholarly communication infrastructures can be leveraged to answer complex user queries (i.e., (meta)data-driven analysis). However, this federated system currently has some limitations: the scope of querying within the federated service is limited; it does not support a wide range of queries, which restricts the ability to retrieve diverse data types or conduct complex data searches across the integrated infrastructures. Furthermore, the current capabilities for filtering content within the federated query service is limited. While scholarly artefacts can be retrieved simultaneously from multiple scholarly infrastructures, the options for filtering these results based on specific criteria are not suficiently comprehensive or user-friendly. Additionally, after retrieving the data, post-processing steps are required to refine the results according to the user’s specific requirements.
To address these limitations, we extend the GraphQL-based federated system and integrate
other scholarly communication infrastructures, namely, ORKG Ask3, OpenAIRE Graph [
Our contributions are as follows: 1. We extend the GraphQL-based federated system to include the ORKG Ask, OpenAIRE Graph, and Semantic Scholar infrastructures, and to retrieve the fragmented scholarly content via a single endpoint to answer complex user queries. 2. We define a clear and straightforward mechanism for filtering results in a user-friendly manner. This enhancement includes filtering capabilities for data obtained from infrastructures beyond ORKG.
In order to demonstrate the practicability of federated scholarly infrastructures, the straightforward implementation and the added value, we propose the following scenarios reflecting diverse information needs of diferent stakeholder groups: • Researchers: Filtering papers based on the specified criteria e.g., find all COVID-19 papers reported 0 estimate less than a threshold. • Funders: Retrieve the most significant papers (both in terms of statistical impact and academic impact) published under a specific grant. This entails identifying research outputs that have advanced the field as well as demonstrated solid results for a particular problem. • Bibliometricians: (1) Create a network of co-authors for a given researcher, focusing on those collaborations that have produced the most significant papers. This involves analyzing the content of researchers articles and identifying those that have reported statistically significant results. (2) Uncover the research focusing on novel problems that have received a high number of citations. This involves identifying such papers and assessing their citation counts.
In the context of above discussion, we thus focus on addressing the following research question:
How can we reuse interoperable scholarly knowledge to support complex (meta)data-driven analysis?
Several services have been developed to process federated queries across diferent databases or
scholarly communication infrastructures. BioThings Explorer [
To access biological data, ROBOKOP (Reasoning Over Biomedical Objects linked in
Knowledge-Oriented Pathways) [
Tong et al. [
DataCite ofers a GraphQL endpoint for the PID Graph, connecting persistently identified resources from DataCite, ORCID, ROR, and other sources, and providing standardized metadata for these resources. Similarly, we implement a GraphQL endpoint to enable access to ORKG content. Additionally, we also integrate the ORKG Ask, OpenAIRE Graph, and Semantic Scholar REST APIs to fetch diverse information regarding diferent scholarly artefacts. ORKG Ask is an advanced search system that helps to find and extract valuable information from a vast corpus of research articles. It revolutionizes the way researchers navigate scholarly literature by leveraging cutting-edge Large Language Model (LLM) technologies to deliver precise and relevant answers according to research queries. By federating also ORKG Ask, the system obtains the ability to leverage generative content, which provides a possibility to address missing data. When certain pieces of information are not available in the ORKG papers, the system can use generative models to predict the missing details, thereby enhancing the comprehensiveness and utility of the search results. OpenAIRE Graph provides metadata about various scholarly artefacts (articles, datasets, software) and other objects, including projects, organizations, and researchers. Semantic Scholar is an academic search engine developed by the Allen Institute for AI. It uses artificial intelligence and machine learning to help researchers find relevant information eficiently.
Our federated query service facilitates cross-walking between metadata about artefacts (such as articles and datasets) with their context (such as people and organizations) and the content of the articles, thereby supporting complex (meta)data-driven analysis. Figure 1 illustrates the enhanced federated architecture. The proposed federated graph plays a crucial role in enabling complex (meta)data-driven analysis. In addition to metadata analysis, infrastructures can utilize article’s content from the ORKG to perform new kind of analysis on data. For example, a researcher may discover all research outputs (papers) published under a particular grant that have impact (high number of citations) as well as have significant results (in a statistical sense of < .001).
A researcher discovered an ORKG COVID-19 comparison for the virus’ basic reproduction (0) estimates. This comparison can be accessed through the federated GraphQL endpoint using its DOI: https://doi.org/10.48366/r44930. The researcher is interested in filtering studies that report 0 estimate less than a specified threshold and utilize the method generalized growth model. While the comparison includes details about methods used to conduct experiments, not all papers include them. We thus leverage ORKG Ask to complete this missing data in a generative manner. The federated endpoint is designed to facilitate such complex queries by executing the relevant components at designated endpoints, thereby enabling the essential metadata-driven analysis for the research. Listing 1 shows the query executed on ORKG and ORKG Ask in a federated manner. ORKG provides the details of a comparison whereas semantic details (methods, and results) of compared papers are retrievable from ORKG Ask. We can also include additional filters to refine the search results, for example search only those studies that have reported results for the European region. For such a scenario, the GeoNames endpoint within the federated service will be queried to retrieve all countries included in the EU region. The results can be filtered by adding the location property to the where clause: property: "location", value: "EU".
A funder seeks to collect all articles that have employed Named Entity Recognition across various domains, particularly those reporting an accuracy exceeding, for example, 65%. This process will allow the funder to understand the domains where NER has been applied, the methodologies used, and their corresponding accuracies. Additionally, the funder aims to determine if a retrieved artefact is peer-reviewed or not. Such a complex scenario can be addressed by querying OpenAIRE Graph, DataCite, and ORKG in a federated manner. The federated query outlined in Listing 2 is first executed on OpenAIRE Graph to gather all papers acknowledging a specific grant (859136 4), then on DataCite to retrieve peer review information, and finally on ORKG to obtain the corresponding machine-actionable contribution descriptions. These papers are then filtered based on the research problem addressed and the reported accuracy. Additionally, the funder can generate a comparison of all research contributions and analyze the significance of the funded work, considering both its impact and statistical significance.
A bibliometrician wants to find all co-authors of a particular researcher who is working on the estimation of COVID-19 reproduction and published highly significant papers i.e., reported average 95% Confidence Interval (CI) less than some threshold. Thus, she performs a network analysis of the researcher’s collaborations and also determines highly significant studies in the epidemiology research. This task relies on the DOI of articles that reported 95% CI values, retrievable from ORKG thanks to machine actionable content representation. The set of articles of a researcher can be retrieved using ORCID ID by leveraging the DataCite services. Finally, the most important articles meeting the criteria (COVID-19 articles reported CI 95% < 4) can be fetched from ORKG. Listing 3 shows the query executed on DataCite and ORKG in a federated manner. Figure 2 shows the co-authors network of a researcher with whom he has published COVID-19 articles.
A bibliometrician is exploring studies that address fairness in machine learning models, with a specific focus on identifying articles that have reported significant outcomes in fairness metrics and are also highly cited. This task involves accessing articles that have reported fairness metrics, available through ORKG. With a relevant set of articles identified, the number of citations for each can be retrieved from Semantic Scholar. Listing 4 outlines the query executed on ORKG and Semantic Scholar in a federated manner. Lines 3-9 in the query are executed on the ORKG API to fetch the papers’ DOIs, and line 12 is executed on Semantic Scholar to retrieve the citations of each paper.
Federated search is a widely used approach to retrieve data from distributed sources. To address our research question, we presented a federated architecture that crosswalks the metadata and data from DataCite, ORKG, ORKG Ask, OpenAIRE Graph, and Semantic Scholar infrastructures. The proposed federated architecture leverages the machine-actionable scholarly knowledge published in the ORKG, the generative content capabilities by ORKG Ask, and classical (bibliographic) metadata to answer complex user queries. Our federated endpoint enables diferent stakeholders to pose queries to meet their complex information needs. We have shown the practicability of federated search by presenting diferent federated queries: (i) retrieving statistically significant research articles (ii) retrieving highly impactful research articles published under a specific grant (iii) supporting bibliometricians in creating network of co-authors who have published articles with significant results.
We aim to develop ORKG Commons (Figure 3), a web-based platform that will enable interactive exploration of data served by the federated GraphQL service presented here. This platform will allow users to seamlessly navigate and analyze interconnected artefacts retrieved from various scholarly communication infrastructures in a federated setup. A key feature of ORKG Commons will be to allow users to write queries in natural language, which will be automatically converted into GraphQL queries to facilitate broader accessibility and ease of use. Additionally, ORKG Commons will feature dynamic facets for content filtering. These facets will be generated based on the metadata and the content of scholarly artefacts —providing users with enriched options to refine their search results. This interface will provide users with appropriate options to refine their search results and facilitate navigation through the vast and interconnected scholarly knowledge available through various scholarly communication infrastructures.
Federated search is a well-established method for retrieving data from heterogeneous and distributed sources. This paper introduces a federated architecture for a system designed to support cross-walks between scholarly metadata and data. The federated system described has the potential to help users in conducting advanced scientific inquiries. By enabling the formulation of complex information needs, this system supports the exploration and analysis of scholarly knowledge published in scholarly articles together with contextual research information. This, in turn, can lead to more insightful, data-driven discoveries across various scientific domains.
This work was co-funded by the European Research Council for the project ScienceGRAPH (Grant agreement ID: 819536) and the German Research Foundation (DFG) project NFDI4DS (PN: 460234259).
Listing 1: Retreiving COVID-19 comparison from ORKG and other semantic details from ORKG
Ask as well as filtering the results according to the specified condition. comparison( doi: "10.48366/r44930" where: [{property: "METHOD", value: "generalized growth model" }]) { doi label contributions { id label paper { doi label } ORKG Askdetails { method } data { propertyLabel label} }, "ORKG Askdetails": { "method": "The study statistically estimated the cCFR and the basic reproduction number using the exponential growth rate of the incidence." Listing 2: GraphQL query executed on OpenAIRE Graph, DataCite and ORKG to obtain the details of papers published under ScienceGraph project (Grant ID: 819536). 24 25 26 27 } } } }, "data": [...] }] id: "819536", where: [ 1 { # OpenAIRE Graph query 2 project( 3 4 5 6 7 ]) { 8 #DataCite Query 9 peerReview { type } 10 papers { # ORKG query 11 label doi 12 contributions { 13 id label 14 15 }} 16 }} 17 18 Result (shortened): 19 { "data": { "project": { 20 "papers": [{
data { propertyId label } { property: "P32", value: "named entity recognition" }
# search for numeric value { property: "P45075", _GTE: 0.65 } Listing 3: Results obtained by querying the PID Graph and ORKG federation about COVID-19 1 Query: 2 { # DataCite query 3 person(id: "https://orcid.org/0000-0001-8722-6149") { id name 4 # ORKG query 5 papers( where: [ 6 { 7 property: "P32" #research problem 8 value: "Determination of the COVID-19 basic reproduction number" 9 } 10 { property: "HAS_VALUE", _LT: 4 } 11 ]) { label id} # showing paper title and doi 12 } } 13 14 Result (shortened): 15 { "data": { "person": { "id": "https://orcid.org/0000-0001-8722-6149", "name": "Shi Zhao", 16 "papers": [{ 17 "id": "R44910", 18 "label": "Estimating the Unreported Number of Novel Coronavirus 19 (2019-nCoV) Cases in China in the First Half of January 2020: 20 A Data-Driven Modelling Analysis of the Early Outbreak" 21 }] 22 } } } Listing 4: Results obtained by querying the Semantic Scholar and ORKG federation about the study Fairness in machine learning and has reported Balanced accuracy > (75%).