Ontotext Knowledge Graph: A story about triples and the desire to connect everything to everything

In 2022 I had the chance to walk the thorny, as I would later find out, road of my PhD thesis talk towards a vision of marketing where we don’t manipulate the marketing mix, but rather manage knowledge.

I did that thanks to Ontotext’s CEO Atanas Kiryakov (by the way also the author of the most cited paper for semantic annotation: Semantic Annotation, Indexing, and Retrieval who invited me to work with them on developing a knowledge graph from marketing content.

Very long chain of Jira stories short, Ontotext Knowledge Graph is a playground that aims to showcase the power of semantic technology and knowledge graphs. We put our knowledge graph technology and text analytics pipelines to work for Ontotext’s marketing content, letting anyone to talk to the content and navigate resources by type, topic, industry and application areas.

Technically, Ontotext’s Knowledge Graph is built of semantically annotated marketing content, stored in GraphDB with a view to supplying websites, chatbots and other data-fed agents with structured data about Ontotext content and the concepts weaving that content, e.g. products, services, capabilities, events, people, technical topics etc.

It is meant to serve internal stakeholders like employees and management by enabling them to find, navigate, utilize, repurpose, and analyze existing content more effectively. We also envisioned it benefiting external audiences such as customers, partners, and investors by making information about Ontotext’s products, services, and technologies easily discoverable, navigable, and explorable.

OTKG comprises custom ontologies (extensions of schema.org in this case) and custom taxonomies. Its knowledge model is stored in GraphDB, leveraging standard tools for knowledge graph management. Ontotext Refine is used to transform structured and semi-structured content into the Resource Description Framework (RDF) format.

Through Ontotext Metadata Studio (OMDS), semantic content enrichment is applied using text analysis based on marketing vocabularies.

The assemblage allows the content to be classified according to the refined knowledge model and entities of interest. The classified content is then exposed via flexible semantic faceted search facilitated by metaphactory, a knowledge graph platform from metaphacts.

More about the technicalities in Krasimira Bozhanova’s post, Ontotext bright Solutions architect: Ontotext Marketing Gets a Boost from Knowledge Graph Powered LLMs 

Thanks to Vladimir Alexiev and his work on the data models, based on content models, as well as on data transformations, the Ontotext Knowledge Graph is also capable of emitting JSON-LD about the pages we have in it, so that we can embed the code in our CMS (WordPress). Here you can see an intricate weave of the Event as a knowledge graph object, also taking into count the fact that Event is also a content type, together with its likes: Webinar description, for example.

The screenshots below show how Ontotext Metadata Sudio automatic annotations are further transformed in JSON-LD.

Were Tim (an imagined webpage visitor/knowledge graph explorer) an external person at Ontotext, he would have saved time and reduced the effort needed to search, find and reuse content, given the knowledge-graph-powered ways of navigating content. As a visitor to Ontotext’s website he will be having meaningful interactions by enjoying semantic search across content, topical pages built depending on a chosen asset or broader query.
Powered by a knowledge graph, a platform with marketing content allows Tim to “talk” to the content, browse it through faceted search and also search for certain keywords and topics, navigating a tree of concepts. Tim is also able to ask an LLM about anything and get answers which are contextually relevant to Ontotext.

You can see for yourself what does all that mean experientially at: https://kg.ontotext.com/. And Conceptually, Peio Popov of Ontotext framed that well, putting a layer of common needs enterprises look for in a system coupled with an LLM:

Instead of an Epilogue: Ontotext Knowledge Graph Powered Search

The story about OTKG ends in a website search and a demonstrator, but is poised to keep going, fractally following what several years ago we mapped as steps to the process of building a knowledge graph.

We are now at the point where we can maximize the usability of the content (or what our data engineers see – the data :-)). In the upcoming months we will be delivering answers website visitors’ questions via OTKG. And surely we are just in the beginning of making our knowledge graph easy to maintain and evolve.

And I still haven’t given up the idea of using WordPress metadata curation processes as a way to infuse human-curated aboutness in our OTKG.

That I call content to the power of ontological engineering. Despite the nearly impossible task of bridging marketing – which is messy by default, with data modelling – and its formalisations and the need for objects and differentiations.
❤️

Let’s see where the next Jira story will take us.

Thanks for reading this story, which is an unplugged version of the Knowledge Graphs stories I present in my book Being Dialogic.