Cigden Z. Gurgur looks at KGs in the context of blockchain. The article begins with background information on how KGs have been used in advanced analytics and their role in helping AI developers. Gurgur then shows how blockchain’s immutability and verifiability offer designers a way to advance KGs to produce more reliable results. The blockchain/KG combination is an ideal one to build more explainable AI systems, she says. Finally, Gurgur explains how KG-enabled information systems can be used in industrial settings to enhance product development lifecycles, improve factory safety, and enhance information systems to the point where employees need less technical knowledge to perform their duties.
Advances in high-tech sensing, the proliferation of electronic manufacturing records and mobile sensors, and the development of the Industrial Internet of Things (IIoT) are causing manufacturing data to accumulate exponentially. Although this data is often stored in heterogeneous formats and distributed, it’s an important source of the information we need to deploy intelligent production management tools. The process involves knowledge extraction and prediction processes using artificial intelligence (AI) models, the success of which is mainly due to advances in machine learning (ML).
ML, a subset of AI, enables learning from observations (data) and experience (repeated training) and is key to transforming large manufacturing data sets (often called “big industrial data”) into actionable knowledge. This is stimulated by large data sets involving various real-world features and an increase of the computational gains generally attributed to powerful graphic-processing cards.
Knowledge graphs (KGs) that leverage AI and ML technology, particularly deep learning, are now being widely studied for use in manufacturing because of their ability to easily handle large amounts of data and model complex relationships.1 Deep learning, a subset of ML, learns without human supervision. KGs are a powerful data science technique created to mine information from diverse data formats.
KGs’ ability to handle connected data and embrace relationships in a flexible, graphic form makes them highly efficient in domains where data structures are constantly changing and evolving, as in manufacturing. Flexible KG schemas can handle dynamic, uncertain variables and quickly encode domain and application knowledge. KGs complement ML methods, enabling accurate data collection that facilitates faster, more precise AI application development.
For example, a recent Technological Forecasting and Social Change article showed how smart manufacturing could use advanced manufacturing technology and data-mining techniques like KGs to improve product quality while shortening production cycles, enhancing production efficiency, and reducing costs.2
The History of KGs in Advanced Analytics
In the last 10 years or so, KGs have emerged as an important area in advanced analytics and the AI domain, helping to connect data sources and solve large-scale enterprise problems. They are at the core of human-facing technologies, such as search, question answering, dialogue, fraud prevention and investigation, product recommenders, and autonomous systems.
In addition to business problems, KGs have been used to solve social problems involving difficult technical challenges, such as human trafficking. A recent article in IEEE Transactions on Big Data described a KG effort that resulted in an effective semantic search engine to assist analysts and investigative experts in the human-trafficking domain.3
Because data sets are so often scattered, AI developers struggle to discover, share, and manage data from different systems in different formats. This requires understanding, structuring, integrating, and verifying data each time new features or applications are built based on it. One of the main benefits of structuring knowledge in the form of graphs (instead of relational databases) is the flexibility of the schema, which can be defined at a later stage and adjusted over time. This allows more flexibility for data evolution and capturing incomplete knowledge.
Building on a storied tradition of graphs in the AI community, a KG can be defined as a directed, labeled, multi-relational graph with some form of semantics integrating diverse data into a common format. KGs provide graph-structured topologies to organize data and can present interlinked descriptions of its entities, including objects, events, situations, and abstract concepts.4
A 2012 Google blog entry is often cited as having sparked KG development.5 In truth, Google revived KG technology rather than inventing it — a great deal of KG research was done in the 1980s. For Google, KG technology was about enhancing search engine performance through information gathered from a variety of sources.
In 2016, researchers Lisa Ehrlinger and Wolfram Wöß proposed a more widely acknowledged definition: “A knowledge graph acquires and integrates information into an ontology and applies a reasoner to derive new knowledge.”6
In this era of information explosion, KGs have tremendous potential to elicit, integrate, process, use, and popularize large data sets embedded in industrial products and services. KGs allow reasoning about the underlying data, provide significant increased precision with information retrieval, and facilitate complex decision making.7
KGs’ underlying structure offers both humans and machines better knowledge comprehension and interpretation.8 Today’s KGs supplement manual knowledge-engineering techniques with crowdsourcing and use ML to significantly increase automation.
Although KGs can improve AI predictions by providing them with knowledge expressed and used by ML methods, most ML models require a set of feature vectors as input. As a result, considerable research has been done to generate “embeddings” from KGs. A KG embedding transforms the nodes and the edges of the graph topology to a numeric feature vector that can serve as a direct input to the ML model.9
AI experts are therefore manipulating structured KGs for deep learning with relational inductive techniques, transferring learning (inter-domain knowledge sharing), and seeking other methods of infusing KG into ML.10 In some cases, KGs have been used to extend existing data models depicted by domain ontologies and establish a new form of advanced analytics that can capture large, semantically interconnected data sets.11
Even though directed labeled graphs represent a common thread linking today’s KGs with early AI semantic networks, there are some important differences in research methodologies and technical challenges. In early AI semantic networks, the emphasis was on complex logical inferencing; modern KGs focus on supporting advanced analytics operations.12
Additionally, early semantic networks were created using top-down design methods and manual knowledge-engineering processes. They never reached the size and scale of today’s KGs. Modern KGs are larger in scale and are constructed using both manual and automated strategies. The vast proliferation of available data and the data-driven nature of today’s ML support a bottom-up methodology for creating KGs.
Unlike rigid relational database structures, KGs’ flexible semantic data layer allows users to perpetually link and network the complex relationships contained within their data platform and external sources without changing the underlying data, thus enriching the data’s semantic meaning.
Enhancing Enterprise KGs with Blockchain
It began as a digital currency technology, but blockchain has rapidly entered virtually every aspect of our lives, from enhancing food safety and preventing medical errors to diamond provenance information disclosure and artwork ownership authentication. Blockchain provides decentralized trust management in chronological, encrypted, chained blocks to store verifiable, synchronized data across peer-to-peer networks.
Blockchains maintain their data integrity, while providing tamper-proof and secure data storage and immutable task execution. Blockchain’s unique tracing ability lets it identify malicious activity on the network. If a malicious user tries to tamper with an enterprise KG, he or she has to access the pointer from the blockchain, so the user’s account can be traced.
The KG/blockchain combination marries integrity with interoperability and interconnectivity. Blockchain’s visibility into the decisions of all AI agents on a KG network makes it difficult for AI agents to modify or refute decisions. Blockchains also let AI agents collaborate to save new decisions on blocks that can be traced back and are therefore resistant to alteration.
A forward-looking article in IEEE Access recommends using crowdsourcing on a blockchain platform to update KG and AI systems with a “trustful” value.13 The research proposes a cutting-edge, decentralized KG construction method using crowdsourcing, with the business logic of crowdsourcing implemented by blockchain-powered smart contracts to guarantee transparency, integrity, and auditability. This technique represents a beneficial tradeoff between the completeness and the correctness of KG, as it takes full advantage of the wisdom of crowds.
Another set of researchers created a visionary framework to enhance KGs with fundamental blockchain concepts, improving the reasoning algorithm with trustworthy and historical knowledge to produce more reliable results.14 The framework includes a verified, trusted state provided by blockchain technology in KGs to help an AI system show why it made a specific decision. Fully provable explanations of AI decisions can be produced by going back in time via blockchain technology.
Having such an integrated system could provide a path to real-time KGs, amalgamating the unmodifiable and accessible history concept and providing verified KGs by blending the concept of digital signatures, which would build a secure connection between KGs and blockchains.
Furthermore, since KGs are designed for complex data and knowledge integration tasks as well as reasoning tasks and do not require hard-coding knowledge into reasoning algorithms, they resolve the scalability challenges in blockchain implementation.
Semantic linking to a data source is one aspect of what is usually called “provenance information.” Provenance tracks the origin of that data and is one form of metadata that is rarely captured in typical relational databases. However, it is relatively easy to capture in KGs.
The convergence of blockchains and KGs allows dynamic enrichment of logic that ends up in a decentralized graph for trustworthy decision making.
Trustworthy & Explainable AI Systems by Blockchain and KGs
Increasing computational power and big data proliferation are driving AI system adoption. Decision support algorithms are carried out by mathematical models (trained using ML techniques) on data collected from past experiences. However, the opaqueness of AI decisions is a major drawback in systems like industrial design, where precision and safe product development are required.
AI technologies that can provide human-understandable explanations for their output or actions are usually referred to as “explainable AI.” End users wonder about the reasoning behind the decisions made by algorithms, and increasing complexity results in a lack of transparency that negatively affects user trust.
The Cambridge Analytica scandal and the 2016 US election disruptions clearly showed why modern ML methods need to be more transparent.15 A number of initiatives have been launched since then, including the US Defense Advanced Research Projects Agency’s Explainable AI program16 and the EU’s Ethical Guidelines for Trustworthy AI.17 Both encourage the design of ethical systems that humans can understand, manage, and trust.
One way to build explainable AI systems is by using KGs. New research in Semantic Web shows how KGs represent a valuable form of domain-specific, machine-readable knowledge.18 The resulting KG-connected or centralized data sets can serve as background knowledge for AI systems to better explain their decisions to users.
Explainable AI is also needed to combat adversarial attacks against ML and deep neural networks that may poison learning or inference processes. These attacks come in a variety of flavors, such as data set poisoning, internal network manipulation, and side-channel attacks. Malicious actors can cause random or targeted misclassifications by manipulating the environment around the AI system, the data acquisition block, or the input samples. The attack can be as simple as adding adversarial noise to input samples and as malicious as incrementally shifting the decision boundaries during the ML training process.
Blockchain technology can be used to produce trustworthy AI requirements to mitigate biases and guard against adversarial attacks.19 With blockchain, explanation systems, including decision outcomes, can be audited in an immutable, tamper-proof, decentralized way that can be traced with high reliability. If any node fails or leaves the chain, the blockchain remains unaffected.
By merging blockchain technology with KGs, we can achieve next-generation industrial information systems for secure data sharing among stakeholders, maintaining data privacy and integrity through data authentication and robust data adaptation. This type of industrial platform would improve trust, elevate scalability, and increase efficiency through multi-party and multi-agent decision-making systems that follow various consensus protocols. It could be used to host a trusted trail of all records used by ML algorithms before, during, and after the learning and decision-making process.
Enhancing Productivity Through KG-Enabled Industrial Product Development: Recent Examples
Demand forecasting and requirement analysis are crucial topics in industrial product development, and they require massive information inputs and robust analytic models to make better predictions. Processing multi-source information and conducting logical knowledge reasoning are two major strengths of KG-enabled information systems. The explainable capability of knowledge reasoning and recommendations enabled by KG are valuable for demand forecasting and requirement analysis, given that stakeholders care more about the insights and logic behind the results than about ordinary point estimates.
Recently, an article in International Journal of Production Research demonstrated KGs’ ability to collect extensive information from online technical forums and portal websites to capture market trends and other events impacting consumer demand.20
Smart Solution Design
Industrial product development requires a high degree of knowledge synthesis and precision specification. KGs’ ability to gather data from multiple sources, usually in different formats, facilitates the creation of easily extendable, flexible data models ideal for made-to-order manufacturing. Several studies have shown how such automated knowledge extraction and fusion improve manufacturing design capacity.21 Furthermore, KGs’ real-time information exchange enables last-minute customer changes, even after production has begun.
KG-based design systems not only automatically save and store final solutions, but also earlier rejected ideas. The solutions and ideas are stored as knowledge in the KG, creating a more holistic knowledge base for the manufacturer that can enhance the product development lifecycle.22
Risk Prediction & Solution Prescription
Recently, researchers in the manufacturing field have attempted to drive industrial services with KG to optimize process safety and product quality. A paper in Systems Research and Behavioral Science proposed an advanced paradigm to apply KGs in smart factories to support safety management in the manufacturing process.23 Researchers proposed KGs as a way to: (1) improve decision making based on problem diagnoses and (2) predict potential risks based on information (e.g., worker location or machine status) and suggest preventative measures.
Similarly, a recent article in Computers in Industry showed how design rules and context information could be combined to build a computable KG, improving computer-aided design and allowing designers to spend time on design rather than looking for design rules.24
Through a better understanding of the relationship between function-behavior-structure and knowledge representation, a KG-based risk prediction and prescription system could prompt smart components to adjust themselves to solve problems.25
In the industrial product lifecycle efforts, there is typically a huge gap between massive heterogeneous knowledge resources in information systems and system users’ limited cognitive ability. Holistic, nonspecific information is either useless or confusing to users. What’s needed is an information system that can dispense the right information at the right moment to those working on specific designs.
With their ability to distill information, KGs can support those designers to better complete creative manufacturing tasks. For example, a KG-based system was able to deliver helpful information in a multi-language environment to employees without a technical background in fashion design manufacturing.26
KG-enabled multi-disciplinary information systems integrated with blockchain technology can facilitate industrial data mining with trustworthy principles. Such systems are capable of producing originative ideas to help users productively and safely complete product development process tasks with increased precision.
Demand forecasting and requirement analysis, smart engineering solution design, automatic risk prediction and prescription, operational maintenance, and information distillation all lead to time and manpower savings.
By leveraging KGs and blockchains, manufacturing enterprises can tap into innovations like explainable AI; reusable semantic data modeling; and scalable, trustworthy, complex-query performance to help accelerate advanced analytics insights and reduce data operations cost.
1 Gordana, Zeba, et al. “Technology Mining: Artificial Intelligence in Manufacturing.” Technological Forecasting and Social Change, Vol. 171, October 2021.
2 Zeba et al. (see 1).
3 Kejriwal, Mayank, and Pedro Szekely. “Knowledge Graphs for Social Good: An Entity-Centric Search Engine for the Human Trafficking Domain.” IEEE Transactions on Big Data, Vol. 8, No. 3, 1 June 2022.
4 Pan, Jeff Z., et al. Exploiting Linked Data and Knowledge Graphs in Large Organisations. Springer, 2017.
5 Singhal, Amit. “Introducing the Knowledge Graph: Things, Not Strings.” The Keyword, Google, 16 May 2012.
6 Ehrlinger, Lisa, and Wolfram Wöß. “Towards a Definition of Knowledge Graphs.” Joint Proceedings of the Posters and Demos Track of the 12th International Conference on Semantic Systems (SEMANTiCS 2016) and the 1st International Workshop on Semantic Change & Evolving Semantics (SuCCESS’16) Co-located with the 12th International Conference on Semantic Systems (SEMANTiCS 2016). CEUR Workshop Proceedings, 2016.
7 Dalle Lucca Tosi, Mauro, and Julio Cesar dos Reis. “Understanding the Evolution of a Scientific Field by Clustering and Visualizing Knowledge Graphs.” Journal of Information Science, Vol. 48, No. 1, 1 February 2022.
8 Sheth, Amit, Swathi Padhee, and Amelie Gyrard. “Knowledge Graphs and Knowledge Networks: The Story in Brief.” IEEE Internet Computing, Vol. 23, No. 4, 17 October 2019.
9 Wang, Quan, et al. “Knowledge Graph Embedding: A Survey of Approaches and Applications.” IEEE Transactions on Knowledge and Data Engineering, Vol. 29, No. 12, 1 December 2017.
10 Hogan, Aidan, et al. “Knowledge Graphs.” ACM Computing Surveys, Vol. 54, No. 4, May 2022.
11 Nanduri, Jay, et al. “Microsoft Uses Machine Learning and Optimization to Reduce E-Commerce Fraud.” INFORMS Journal on Applied Analytics, Vol. 50, No. 1, 24 January 2020.
12 Chaudhri, Vinay K., et al. “Knowledge Graphs: Introduction, History, and Perspectives.” AI Magazine, Vol. 43, No. 1, 31 March 2022.
13 Wang, Shuai, et al. “Decentralized Construction of Knowledge Graphs for Deep Recommender Systems Based on Blockchain-Powered Smart Contracts.” IEEE Access, Vol. 7, 19 September 2019.
14 Bellomarini, Luigi, et al. “Blockchains as Knowledge Graphs — Blockchains for Knowledge Graphs (Vision Paper).” Proceedings of the International Workshop on Knowledge Representation and Representation Learning (K4RL) Co-located with the 24th European Conference on Artificial Intelligence (ECAI 2020). CEUR Workshop Proceedings, 2020.
15 Hern, Alex. “How Social Media Filter Bubbles and Algorithms Influence the Election.” The Guardian, 22 May 2017.
16 Gunning, David, et al. “DARPA’s Explainable AI (XAI) Program: A Retrospective.” Applied AI Letters, Vol. 2, No. 4, 4 December 2021.
17 ”Ethics Guidelines for Trustworthy AI.” European Commission, 8 April 2019.
18 Lecue, Freddy. “On the Role of Knowledge Graphs in Explainable AI.” Semantic Web, Vol. 11, 2019.
19 Salah, Khaled, et al. “Blockchain for AI: Review and Open Research Challenges.” IEEE Access, Vol. 7, 1 January 2019.
20 Wang, Zuoxu, et al. “A Graph-Based Context-Aware Requirement Elicitation Approach in Smart Product-Service Systems.” International Journal of Production Research, Vol. 59, No. 2, 18 December 2021.
21 Yuan, Jianbo, et al. “Constructing Biomedical Domain-Specific Knowledge Graph with Minimum Supervision.” Knowledge and Information Systems, Vol. 62, 23 March 2019.
22 Wang, Ru, et al. “A Process Knowledge Representation Approach for Decision Support in Design of Complex Engineered Systems.” Advanced Engineering Informatics, Vol. 48, April 2021.
23 Liu, Zimei, et al. “A Paradigm of Safety Management in Industry 4.0.” Systems Research and Behavioral Science, Vol. 37, No. 4, 25 June 2020.
24 Huet, Armand, et al. “CACDA: A Knowledge Graph for a Context-Aware Cognitive Design Assistant.” Computers in Industry, Vol. 125, February 2021.
25 Shi, Hui-Bin, et al. “An Information Integration Approach to Spacecraft Fault Diagnosis.” Enterprise Information Systems, Vol. 15, No. 8, 2021.
26 Peroni, Silvio, and Fabio Vitali. “Interfacing Fast-Fashion Design Industries with Semantic Web Technologies: The Case of Imperial Fashion.” Journal of Web Semantics, Vol. 44, May 2017.