An Introduction to Knowledge Graphs
Understanding how to work with knowledge graphs can give data scientists the ability to not just extract interrelated facts and assumptions from massive collections of data, but can also help in understanding how to form contextual connections and understanding from data via linking and semantic metadata which helps provide a unified approach to data analytics and intelligence. This article is an introduction to knowledge graphs, important concepts and their applications.
Introduction
Knowledge graphs are used to map the collection of data from different sources and create a connection between the different entities in a given subject matter such as people, places or things to provide meaning to the data and remove any Semantic ambiguity (Ristoski and Paulheim, 2016). Graph technologies and analytics offer tools for a system driven by connections — from social connections to financial connections, it helps move data analytics and intelligence from just tabulating and retrieving data to forming real-world meanings and relationships (Needham and Hodler, 2019).
Knowledge graphs can be considered as a type of semantic network which forms the basis of associations between concepts and entities in a network — connections between real-world objects, events, situations or abstract concepts (‘Semantic network’, 2021). Although they are used quite interchangeably, knowledge base and knowledge graphs are not the same with the fact that information in knowledge graphs are organized using graphs which is not true for all knowledge bases. There have been different versions of knowledge graphs proposed by different organizations with each version having varying levels of complexity such as Google’s Knowledge Graph, Knowledge Vault, Microsoft’s Satori, Facebook’s Entities Graph, etc.Image by author
In 2012, Google introduced its knowledge graph called “Google Knowledge Graph” harnessing approaches that have been around since the beginning of modern AI to enable its users to discover new information quickly and relevant to their queries (Singhal, 2012).
Resource Description Framework and Ontologies
Knowledge graphs are a combination of different paradigms, a database to provide a way to explore the data through structured queries, a graph to provide a network data structure, a knowledge base to interpret the data and infer new facts and an ontology to serve as a formal semantic data models used to define the types of objects that exist within the subject matter and the properties used to describe them (Ontotext, no date).
Resource Description Framework
Image by authorThe Resource Description Framework is a graph data schema used for describing the semantics, or meaning of information in the form of triples which are based on an Entity Attribute Value (EAV) model, a combination of a Resource, a Property, and a Property value (subject, predicate, object) — forming a Statement used to represent complex relationships between different resources (Wu et al., 2014). This means RDF gives us the opportunity to infer from data by discovering new facts about the data based on a set of rules and due to the semantic nature of RDF we do not need to have explicit relationships stored which means RDF can generate new facts from existing triples.
In Table 1, “Muizz Studies Machine Learning Engineering”, Muizz is the subject, Studies is the predicate, and Machine Learning is the object. It is designed to provide an understandable bit of information to computer applications with pre-created joins compared to RDBMS as seen in the example.
The RDF schema in Figure 1 is everything connected to the rdf:type data science, we see that the entities pointing to data science is a topic under data science and we see that project management although have similar characteristics to Data Science because they both take courses in project management they are not part of the Data science rdf:type and as such are not data scientists. This makes it easy to see connections between entities without having to create complex joins since we know any entity that points to data science is a part of the class. RDF provide interoperability through a uniformImage by author structure that supports the reuse of existing standards and ontologies. They use standards that are unique across different systems which facilitate data integration and publishing by enabling different agents, services and applications to share knowledge and intelligence (Noy et al., 2005).
Ontologies
In computer science ontology is how we formally create conventions for naming and definition of the categories, properties and relations between domains of discourse (‘Ontology (information science)’, 2021). Ontology is different from taxonomy which is a controlled set of vocabulary used to make it easier to find, manage and present related information — they express hierarchical relationships within a category (SCHWEIZER, no date).
Ontology is a type of knowledge base which helps build relationships between people, place and things i.e classes, entities and properties, enabling users to link multiple concepts to other concepts in a variety of ways. Ontological representations allow semantic modelling of knowledge, this means it adds a basic meaning to the data and the relationships that lie between them and are used to form the knowledge bases in artificial intelligence (AI) applications (SCHWEIZER, no date).
Knowledge graphs can be described as a very large ontology with additional features such as data collection, extraction, integration of data from external sources which extends a knowledge-based system and a reasoning engine (Ehrlinger and W oß, 2016). Languages such as Web Ontology Language (OWL) are used to express ontologies, they are designed to represent extensive knowledge about objects and the relations between them by specifying object classes and relationship properties as well as taxonomies (Ontotext, no date).
Usability of knowledge graphs
We will be exploring the use of knowledge graphs for analytics and business intelligence. Telecommunication is all about connections — interchanging information between people, satellites or computers, this makes knowledge graphs a perfect candidate for modelling, storing and querying telecommunications data of all kinds. Telecommunication systems have architectures combining various complex network structures and systems, they provide a diverse range of product offers and bundles and are highly invested in customer satisfaction and churn rate in a highly competitive landscape, implementing a knowledge graph would not only provide an efficient means of monitoring the network systems itself but also enable the business to be more agile in dealing with their customer needs and requirements.
Consider a large telco with millions of customers, agents and partners. They use monitoring systems to monitor customer usage, key performance indicators, application operation and Generation of reports and event notifications (SLAC,2021). Such organization would typically have analytics questions such as:
Knowledge graphs reduce the cost of some expensive joins used in traditional databases as it requires lesser resources to answer analytics questions. In traditional business intelligence platforms finding connections between different relational databases requires time-intensive data modelling and query operations. With each new business question that arises, we would require a new dataset and schema which is not a sustainable approach to intelligence and knowledge (Clark, 2020). Implementing a customer intelligence knowledge graph that focuses on the most customer-relevant entities, such as most common among its users: bundle types, complaints, data usage and subscription rates would give a competitive advantage to the business. Focusing on those domains that have the greatest chance of delivering the best and delightful user experiences would help improve analytics and intelligence in such organizations (Kempe, 2014).
Sample Real-world Knowledge Graph: Microsoft Academic Graph
The Microsoft academic graph is a knowledge graph implementation of academic information and data — it has a collection of entities such as people, publications, fields of study, conferences, and locations. It provides connections between researchers and research related to them which might have been difficult to determine (Noy et al., 2019).
Microsoft Academic Knowledge Graph (MAKG) is a large RDF data set with over eight billion triples of information about scientific publications and related entities, such as authors, institutions, journals, and fields of study. MAKG uses data licensed under Open Data Commons Attribution License (ODC-By). This knowledge graph is able to bring together large volumes of scholarly publications and data to the web by using RDF dump files, data sources from the Linked Open Data cloud with URIs that works and connect to multiple data sources (Farber, 2006). Microsoft academic graph is centred around the acquisition of knowledge from text through the identification of lexical constructs of semantic objects which represent entities or connections. MAKG uses a hierarchical entity type called concepts to represent the semantic contents of a document, it is an abstract way to define entities such as authors and affiliations and as such no concrete way of defining them (Kuansan et al., 2019).
References
[1] Clark, K. (2020) What is a Knowledge Graph?. Available at:https://www.stardog.com/blog/what-is-a-knowledge-graph/ (Accessed: 11 May 2021).
[2] Ehrlinger, L and W¨oß, W. “Towards a Definition of Knowledge Graphs.” SEMANTiCS (2016).
[3] Farber, M. (2006) The Microsoft Academic Knowledge Graph: A Linked Data Source with 8 Billion Triples of Scholarly Data. Institute AIFB, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany.
[4] Feilmayr, C and W¨oß, W. “An analysis of ontologies and their success factors for application to business Data Knowledge Engineering.” SEMANTiCS (2016).
[5] Kempe, S. (2014) Why Your Business Needs A Customer Data Knowledge Graph?. Available at:https://www.dataversity.net/business-needs-customer-data-knowledge-graph/ (Accessed: 11 May 2021).
[6] Kuansan, W., Zhihong, S., Chiyuan, H., Chieh-Han, W., Darrin, E., Yuxiao, D., Junjie, Q., Anshul, K., Alvin, C., and Richard, R.(2019) ‘A ReOntologiesview of Microsoft Academic Services for Science of Science Studies’, Journal of Frontiers in Big Data, (2), pp.45. Available at: https://www.frontiersin.org/article/10.3389/fdata.2019.00045 . (Accessed: 11 May 2021).
[7] Noy, N., Gao, Y,, Jain, A,, Narayanan, A,, Patterson, A, and Taylor, J, (2019) Industry-Scale Knowledge Graphs: Lessons and Challenges. Available at: https://cacm.acm.org/magazines/2019/8/238342-industry-scale-knowledge-graphs/ (Accessed: 10 May 2021).
[8] Noy, N., McGuinness, D., and Hayes, P.J. (2005) ‘Semantic Integration Interoperability Using RDF and OWL ’, W3C Journals. Available at: https://www.w3.org/2001/sw/BestPractices/OEP/SemInt/ (Accessed: 11 May 2021).
[9] ‘Ontology’ (2021). Wikipedia. Available at https://en.wikipedia.org/wiki/Ontology. (Accessed: 11 May 2021).
[10] ‘Ontology (information science)’ (2021). Wikipedia. Available at https://en.wikipedia. org/wiki/Ontology_(information_science). (Accessed: 11 May 2021).
[11] Ontotext (no date) Available at: https://www.ontotext.com/knowledgehub/fundamentals/whatis-a-knowledge-graph/ (Accessed: 11 May 2021).
[12] Ontotext (no date) Available at: https://www.ontotext.com/knowledgehub/fundamentals/whatare-ontologies/ (Accessed: 11 May 2021). (Ontotext, no date)
[13] Needham, M. and Hodler, E. A. (2019) Graph Algorithms: O’Reilly Media, Inc.
[14] Ristoski, P. and Paulheim, H.(2016) ‘Semantic Web in data mining and knowledge discovery: A comprehensive survey, Journal of Web Semantics’, Journal of Web Semantics, 4(2), pp.1–22. Available at: https://doi.org/10.1016/j.websem.2016.01.001. (Accessed: 11 May 2021).
[15] SCHWEIZER, C. (no date) What is the Difference between Taxonomy and Ontology? It is a Matter of Complexity. Available at: https://www.earley.com/blog/what-difference-betweentaxonomy-and-ontology-it-matter-complexity (Accessed: 11 May 2021).
[16] ‘Semantic network’ (2021). Wikipedia. Available at https://en.wikipedia.org/wiki/ Semantic_network. (Accessed: 11 May 2021).
[17] SLAC National Accelerator Laboratory (2021) Available at: https://www.slac.stanford.edu/xorg/nmtf/nmtf-tools.html (Accessed: 11 May 2021).
[18] Singhal, A. (2012) Introducing the Knowledge Graph: things, not strings. Available at: https://blog.google/products/search/introducing-knowledge-graph-things-not/ (Accessed: 11 May 2021).
[19] Stanford CS 520 (no date) Available at: https://web.stanford.edu/class/cs520/2020/ notes/What_Are_Some_High_Value_Use_Cases_Of_Knowledge_Graphs.html (Accessed: 11 May 2021).
[20] Wu, B., Zhou, Y., Yuan, P., Jin, H., and O’Hara, J. (2014) ‘SemStore: A Semantic-Preserving Distributed RDF Triple Store.’, In Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management (CIKM ’14). Association for Computing Machinery, New York, NY, USA, Available at:https://doi.org/10.1145/2661829.2661876 (Accessed: 11 May 2021).
