Multiple Knowledge GraphDB (MKGDB)

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2325–2331 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2325 Multiple Knowledge GraphDB (MKGDB) Stefano Faralli1, Paola Velardi2, Farid Yusifli2 1 University of Rome Unitelma-Sapienza, stefano.faralli@unitelmasapienza.it 2 University of Rome Sapienza, velardi@di.uniroma1.it,yusifli.1772848@studenti.uniroma1.it Abstract We present MKGDB, a large-scale graph database created as a combination of multiple taxonomy backbones extracted from 5 existing knowledge graphs, namely: ConceptNet, DBpedia, WebIsAGraph, WordNet and the Wikipedia category hierarchy. MKGDB, thanks the versatility of the Neo4j graph database manager technology, is intended to favour and help the development of open-domain natural language processing applications relying on knowledge bases, such as information extraction, hypernymy discovery, topic clustering, and others. Our resource consists of a large hypernymy graph which counts more than 37 million nodes and more than 81 million hypernymy relations. Keywords: knowledge graphs, graph databases, hypernymy graphs 1. Introduction Since the end of the last century, both semantic networks (Allen and Frisch, 1982) and knowledge graphs (KGs) are playing the important role of representing entities and re- lations with high reliability, explainability, and reusability (Bonatti et al., 2018). KGs have had an impact on the development of many fields of research (Camacho-Collados et al., 2018) ranging from the Semantic Web (Shadbolt et al., 2006) to Natural Lan- guage Processing (NLP)1. In order to speed up and support the development of novel knowledge-based applications, we present MKGDB, a (very) large scale graph database created as a combination of multiple taxonomy backbones, extracted 2 from 5 existing knowledge graphs namely: Con- ceptNet (Speer et al., 2017), DBpedia (Lehmann et al., 2015), WebIsAGraph (Faralli et al., 2019), WordNet (Fell- baum, 1998) and the Wikipedia hierarchy of categories. The resource combines multiple lexical knowledge graphs representing entities and hypernymy relations, both auto- matically harvested from corpora (i.e., WebIsAGraph) and crafted by human experts (like WordNet and others). Thanks to the availability of different streams of knowl- edge and the versatility of graph database technologies, we exploit methodologies leveraging topological features from multiple knowledge graphs. As an example, MKGDB contains a high number of cross- link edges (edges connecting nodes belonging to different knowledge graphs) which are fundamental information in algorithms such as linking and mining heterogeneous data (P and Jurek-Loughrey, 2018), entity alignment between knowledge graphs (Trisedya et al., 2019) and noisy graph pruning (Faralli et al., 2017). Our resource is compiled on top of the Neo4j platform3. The Neo4j, in comparison to other state of art technologies (see Section 2. for a review on graph database systems), of- fers both a scalable solution enabling to manage large scale graphs and specific textual based indexing functionalities to 1Michael Galkin, Knowledge Graphs in Natural Language Processing @ ACL 2019, https://towardsdatascience.com/ knowledge-graph-bb78055a7884 2at time of writing 3https://neo4j.com/ better support natural language processing applications. Other filed of applications and studies that might benefit from the availability of MKGDB are, for example: • applications aimed at generating graph embeddings (Wang et al., 2017) where combining …

Content truncated to 500 words