Ontology Patterns Bring Order to Knowledge Graphs

August 02, 2019 by Stefan Summesberger

SEMANTiCS 2019 Keynote Speaker Valentina Presutti coordinates the Semantic Technology Laboratory of the National Research Council (CNR) in Rome. She received her Ph.D in Computer Science in 2006 at University of Bologna (Italy). She has coordinated, and worked as researcher in, many national and european projects on behalf of CNR and she co-directs the International Semantic Web Research Summer School (ISWS). Valentina serves in the editorial board of top journals such as Journal of Web Semantics, Journal of the Association for Information Science and Technology, Data Intelligence Journal, Intelligenza Artificiale. She’s been involved in many research projects. To name a few: MARIO has developed social robotics technologies for eHealth, IKS contributed to create semantic tools for content management systems, NeOn has introduced theoretical foundation and practical resources for ontology design patterns.  She has 100+ publications in international journals/conferences/workshops on topics such as semantic web, knowledge extraction, and ontology design. Her research interests stand at the crossing between semantic web and artificial intelligence and include knowledge graphs, empirical analysis of the semantic web, and social robotics. In this Interview Valentina talks about ontology design patterns and knowledge graphs.

Earlier this year you talked about engineering ontologies with ontology patterns. Please elaborate on this. What fascinates you about this area? What should we be on the outlook for in the near future? 

In general terms, my “thing” for patterns has to do with the aesthetics of a complex system. Patterns allow a complex system to be modularised and self-explaining, hence more (re)usable. This is true in architecture and in software engineering, where patterns provide solutions to recurrent problems and form a language for supporting communication among designers. Knowledge graphs are complex systems and ontologies are a means to govern them as far as their organisation, semantics, and interoperability are concerned. We are used to hear that knowledge graphs are valuable key assets for building intelligent systems. This is true under the assumption that they are interoperable, but when using them in real-world cases, they show their wild nature. Ontology design patterns play a crucial role for enabling such interoperability, however there is still lack of proper tools for widely spreading their adoption. Research-wise, this area offers a broad range of interesting problems, e.g. detecting and matching ontology design patterns in existing knowledge graphs, formalising relations between ontology design patterns, refactoring ontologies based on patterns, pattern-based reasoning, just to name a few. 

What are the main benefits of engineering ontologies with ontology patterns? Could you boil this down to a list of 5?

  1. Maximising interoperability between ontologies
  2. Easing ontology reuse and linking  
  3. Maximising ontology quality 
  4. Developing ontologies with a sound methodology
  5. Supporting collaborative design (as they provide a common language)

You are an expert on social robotics as well. What are the core concepts and drivers behind this idea?

Social robotics can be described from different perspectives. The main idea behind it is to develop robots that feature socially plausible behaviour as opposed to service robots that are meant to perform specific tasks e.g. scanning the terrain on Mars or vacuum cleaning our apartments. From a technological perspective social robots can be life-changers, especially for people with disabilities and considering that our life expectation is growing, eventually all of us may benefit from them. Scientifically, what's really interesting in social robotics is that by implementing and testing computational versions of psychological, cognitive and social theories we understand their limits, possibly prove their soundness, and ultimately better understand ourselves.

Discuss the potential of Social Robotics with Valentina. Register for SEMANTiCS 2019!

You also are a profound expert on common sense knowledge and personalized knowledge graphs. What are these concepts about and how do they connect with the other areas you research in?

With “personalised knowledge graphs” I mean those knowledge graphs that include your own personal data. They are important in social robotics, when the robot has to act as a companion, especially if you’re an elderly with some issues with your memory. You want the robot, for example, to know about you, to remind you about your past experiences, your friends and family, and keep track of the main events of your day, in case you need to refresh them.

Common sense knowledge is the knowledge we all share and use for every communication and task we perform. This knowledge is never explicit, however if we interact with another person, even if we never met her before, we make a number of assumptions on such a shared knowledge. Similarly we assume that we are interacting with an intelligent agent, able to perform reasoning that relies on such shared knowledge. Sharing common sense knowledge between humans and machines, and for machines being able to use common sense reasoning are two main challenges in AI. Common sense is of course very relevant for social robots. The connection between knowledge graphs and common sense lies on how to represent common sense. Ideally, the Web and more specifically the web of data provides a potential source of global knowledge for artificial agents. Nevertheless, the current situation shows a huge amount of structured and formalised data that suffer from lack of interoperability and incompleteness, this is a showstopper when extracting/deriving common sense knowledge. My research on ontology design approaches the interoperability problem and developing applications for social robots allows me to formulate questions related to the suitability of linked open data as background knowledge for intelligent agents. Related to this, I am now working on performing large scale empirical observations on LOD, from a knowledge engineering perspective. I am trying to answer questions such as how ontologies are actually modelled, how to distinguish common sense facts from domain-specific facts, to what extent LOD contains (explicitly or implicitly) common sense knowledge.

Of the areas discussed above, what are the most exciting projects and applications on your radar in this field and what do they do?

Recently, we have released a new resource named ArCo: the knowledge graph of the Italian Cultural Heritage (see also the resources paper that will be presented at ISWC). ArCo is exciting not only because it brings huge linked open data on the Italian Cultural Heritage on the Web. It also brings, to knowledge graph creators, an exemplary knowledge graph project: ArCo has been developed by following — and experimenting with — pattern-based design, by involving actors from different organisations and with very different goals. It is accompanied by a rich documentation set including an ontology test suite, annotation of used patterns, competency questions, etc. The project also features the development of a set of prototypes for supporting pattern-based design.

I am also involved in a new social robotics project that aims at building an artificial companion that can help elderly in keeping their autonomy as long as possible. I believe that this is an extremely important and exciting area of research. It both allows to approach open research challenges, and have a concrete impact on our society. 

I am also enthusiastic about analysing LOD with the idea of “discovering” properties of knowledge graphs at large. I have been cultivating this idea since long: one of my papers analyses Wikipedia links to observe emerging patterns that describe specific categories of entities. Wikipedia is a very large collaborative resource, hence studying it is certainly interesting, but with the whole LOD, we have a “gold mine” for such analyses! A couple of years later, I attended a very inspiring talk by Frank van Harmelen at one edition of the International Semantic Web Research Summer School. In simple terms, he suggested to study LOD using a natural science approach: observing properties of data as if they were natural entities. What I am working on lately is observing what LOD looks like from a knowledge engineering perspective: a recent collaboration between my group and Frank's on this topic is reported in a research paper that will be presented at ISWC in October.  I think that such studies will help us improving knowledge engineering methodologies, tools and understanding whether and where we got it wrong in the past.

Which exciting use cases did you recently encounter in these fields?

Working with data from the Public Administration may sound less exciting than social robots but in fact when, as in my case, you have the chance to collaborate with an agency that deals with all possible data and has the goal of integrating it, it may be quite challenging and interesting at the same time. My group has contributed to the development of OntoPiA [7] the reference network of ontologies and controlled vocabularies for the Italian PA. The real challenge here is to have a network of ontologies able to support a knowledge graph covering diverse domains such as IoT, tourism, public services, transportation, etc. This complex use case pushed us in adopting a layered ontology architecture where ontology patterns and foundational ontologies come into play for maximising interoperability. What’s exciting is that they are actually collecting significant amounts of data and that adoption from local PAs is gradually growing. This not only may lead to an increasing amount of new and more effective and intelligence  services built on more and more connected data, but also to achieve a better understanding of Italian PA's semantics and processes. 

Referring to what I said before, cultural heritage and assistive robotics are also exciting domains to work with. ArCo’s guiding use cases are different but there are two things in my opinion that makes it special: one is that it models one of the biggest and most important cultural heritages of the world, the other is that, differently from most existing cultural heritage knowledge graphs, ArCo ontologies model the domain, not only the metadata of cultural properties. A mid-long term use case is to link ArCo’s data with those about studies, research, restoration, etc. of cultural properties as well as those coming from different knowledge bases, for example about legal procedures such as confiscation. We are working on this and the result may have a significant impact as a resource for researchers in the humanities. 

One of the most engaging and moving use cases I have worked on is with elderly affected by dementia. The goal here is to help them slowing down the degeneration of cognitive abilities, for example their memory. I’ve had the opportunity to meet some people suffering from this disease and willing to share their experience. It was very touching, they just want to be helped remembering because of the profound sadness of being unable to remember the nice time with their loved ones. It’s like, they say, if you lose a piece of your life with them, and you want it back. Based on input from psychologists and physicians, we have built a personalised knowledge graph, connected with multimedia content such as photographs and music tracks that the robot uses for interacting with its human and helping her remembering places, people, events, etc. Although there’s a long way to go before this type of applications can work with a smooth interaction, the prototype we developed has been welcomed with unbelievable enthusiasm. It has been the first time in my life that I realised how my work can really make a difference and be life changing, and more importantly, bring happiness to people that may have thought it was not possible anymore.

Your SEMANTiCS 2019 Keynote will provide insights in such areas as ontology design, knowledge graphs, cultural heritage, eHealth and assistive robotics. What can we expect? Why must we not miss this talk?

I admit it was not easy to decide the focus of my talk. I have been unsure whether to talk more about my recent work on observing LOD at large scale or to tell a story on ontology design and point out current challenges. I decided to focus on knowledge graph design, which is the unifying topic of my research. This gives me the opportunity to both give an informative talk and also to connect to some results and challenges from social robotics and empirical knowledge engineering, where knowledge graphs and their design are protagonists anyway.  You don’t want to miss this talk if you want to get inspired and become more knowledgeable for your next knowledge graph project.

Do you want to attend this talk? Register for SEMANTiCS 2019!