As knowledge graphs have made their way into more and more areas of our lives through smarter search and immersive information navigation, they are also gradually turning into the default way we expect to interact with knowledge. When we search for people, documents, places, and events, we are now able not only to use a bunch of keywords but to also ask questions and receive answers about a given subject of interest. Take Google’s knowledge graph, underpinning Google’s search, we can use it to ask when a historical figure was born, which cities hosted the Olympics, how to get to the nearest pizza, and so on.
But what about questions related to intangibles?
Imagine being able to search for information about how London smelled back during the days of the Industrial Revolution? Or what does the Battle of Waterloo smell like? Or to browse repositories by asking questions like “What were the new odors that appeared during the Industrial Revolution?” or “What were the most frequent smell sources in London in the 18th century?”. Or be even braver and ask what’s the smell of love?
What if we told you could?
Enter Odeuropa.
Nose-first data: the project Odeuropa preserving the smells heritage of Europe
No doubt our olfactory perception contributes to and enriches our understanding and experiencing of the world. The question is how we can use data and cutting-edge technology to learn about the smell experiences and the perception of different smells in the past.
The answer is through the wealth of object, text, and image data, which serve as a reference to olfactory experiences. It’s these data that the Odeuropa project aimed to tame by developing methods for identifying and tracing olfactory reference and also by designing a data model to represent olfactory knowledge.
In more detail, Odeuropa is an EU-funded project aiming to explore and codify olfactory descriptions, experiences, and memories in European history. Applying state-of-the-art AI techniques to textual and image cultural heritage data, the project tries to enable the identification and traceability of smells. Being an interdisciplinary endeavor, Odeuropa bundles expertise in sensory mining and olfactory heritage. It is home to research and development of novel methods to help scientists collect information about smell from texts and images.
After processing text and image datasets spanning centuries of European history (1600-1920), the project enabled some interesting applications. Among them are various open-source tools and demonstrators to help the navigation of the wealth of smell-related data, along with open-access corpora, vocabularies, and benchmarks available on Github. In addition, there is open access to the Odeuropa data model, as well as a Historical Scent Collection website. Last but not least, thanks to the project we can enjoy an intriguing searchable website that enables us to discover the smells of the past – The Smell Explorer. By July 2023, it had already held nearly 800.000 olfactory depictions and descriptions.
And all that feast of ways to see, touch, and conceptualize smells is technically powered by an even more exciting work: the European Olfactory Knowledge Graph (EOKG).
European Olfactory Knowledge Graph (EOKG): The data language of smells
Knowledge graphs, combined with semantic technology, provide methods and standards for integrating different data sources. That said, to digitally preserve and further promote and present olfactory heritage, the Odeuropa team curates multi-modal information, following semantic web standards, and stores the enriched data in EOKG. It allows the storage and access of information about the chemical composition of a smell, the structured description of its perception by a person, data about the odorant object or place, and the contextual spatiotemporal information can coexist in the same graph.
In the building of EOKG, researchers relied upon and extended existing ontologies as well as introduced classes and relations specific to olfactory and perception in general. They developed, specialized vocabularies in machine-readable formats, where each term could group cross-language synonyms and had explicit logical relationships with other terms. For example, a neroli scent can be part of the fruity scent family, and it was historically believed to help combat the plague. (You can read in detail how the EOKG was built in: the European Olfactory Knowledge Graph, version 1.)
EOKG not only organizes the sparse olfactory information, captures the involved logical relationships, and enables their access, search, and reuse. It also provides specialized knowledge to feed AI applications. As the researchers behind the EOKG put it:
Leveraging the data to automatically group similar odours, infer the dominant olfactory features of a particular time period or a place, answer smell-related questions, and improve the search of olfactory references in texts and images
Lisena, P., van Erp, M., Bembibre, C., and Leemans, I. (2021).
Data Mining and Knowledge Graphs as a Backbone for Advanced Olfactory Experiences.
In more technical detail, the knowledge graph infrastructure includes a GraphDB database, a SKOSMOS instance, a Web API, and a repository containing dumps and scripts, forming the ‘data backbone’ of the Smell Explorer. It can be queried through a SPARQL endpoint, as described in Capturing the Semantics of Smell: The Odeuropa Data Model for Olfactory Heritage Information.
Tripling the intangible: Ontotext GraphDB powering the European Olfactory Knowledge Graph
Apart from being an example of innovative interdisciplinary collaboration, Odeuropa is also yet another cool project using Ontotext GraphDB. It powers the exploratory search engine and the other tools we mentioned.
Currently, Raphael Troncy and the team are hosting the Odeuropa knowledge graph on an enterprise-grade GraphDB having 472,463,753 total statements, of which 103,375,793 are explicit and 369,087,960 are inferred.
The first version of the EOKG was hosted on a GraphDB instance on EURECOM servers, accessible at http://data.odeuropa.eu/. As the authors share, they chose GraphDB as “one of the most popular graph databases available on the market” because of the following benefits:
- GraphDB offers a free-to-use solution
- GraphDB supports RDF star
- GraphDB exposes a web server that can be used for exploring/querying a database and dereferencing URIs;
- GraphDB provides visual interfaces for aggregated data.
All the Odeuropa data stored in GraphDB are accessible via client API accepting SPARQL queries and also through a SPARQL endpoint, exposed in the UI at http://data.odeuropa.eu/sparql. The endpoint provides data in different formats, including JSON, XML, CSV, and TSV. URI dereferencing has been put in place using internal URI rewriting functions based on the ProxyPass feature of the Apache Server. In addition, an interactive visualization of the graph is available through GraphDB’s feature for rendering results as a visual graph.
Epilogue: Follow Your Nose, literally!
EOKG is a gorgeous showcase of the power of knowledge graphs to provide the technological backbone of rich navigation through heterogeneous resources. It is yet another evidence that a little semantics goes a long way. Making the proverbial Semantic Web (FollowYourNose) literally possible, the Odeuropa project is an example of the riches semantic technology brings: unique search experience, multi-modal exploration of resources, and integration with a large variety of public domain resources.
Thanks to semantic technology, and GraphDB in particular, bolstered up by the scientific acumen of Raphael Troncy and team, we can actually ask “What’s the smell of love?”. And, the answer we get is richly layered with plenty of referenceable cultural data.
Ready to follow your nose walking through your data?
 |
Ontotext’s GraphDBGive it a try today!
|
Share this article
