Visual Knowledge Graph — make sense of things you never heard
In occasion of Wikimania 2016, held in my native Country, I want to share my side-project: a research on knowledge graphs to query context of topics and pathways between topics.
It brings knowledge discovery to Wikipedia.
I applied my research and learning on knowledge graphs; UX and material design UI; full-stack development and architecture design.
In between of human memory and memory of humanity
The idea behind this project is a quest for the link between human memory and collective memory. By human memory, I mean a network of experiences and concepts that we recall to interprete new things we learn. By collective memory, I mean a sort of memory of humanity: a network collecting different domains expertise.
Information is growing (see a great book of Cezar Hidalgo for a physics approach to it) and so do knowledge structures, that are becoming more and more complex.
The seminal step is to built a first brick of collective memory, that, on average, can provide results people expect in their own field, and provides for meaningful context and suggestions when they have to research and instantly make sense of new stuff.
Nifty.works is about untangling knowledge structures: you can ask “Tell me about CRISPR, academic papers and companies working on it”, you can ask “Tell me about Science and Democracry” or “Science” and “Free will”; you can ask for which investors are similar respect to the companies they invested in and the patents holded by those companies; you can ask for knowledge embedded in patents to make products — and you get answers tailored to your background.
Well, not so fast, but this is how I imagine a utility to help people learn faster, to make sense of interdisciplinary disciplines, to allow also laypeople have right to access premium knowledge structures (see the brilliant model of sci-hub) and ultimately automate creative production of new knowledge— explore the patterns of how we acquire new knowledge, recombined it with our personal memory, and form “mind-maps” of a subject that make sense to us but are also understandable by others.
Here some first experiments in suggesting answers for complex questions:
I am basically applying betweeness centrality algorithms over weighted knowledge graphs, attempting to find meaningful connections between topics.
Would you help in tuning or make a smarter approach by integrating in-depth knowledge graphs?
What I learnt
In this project I experimented with algorithms, UX and learnt full-stack development building a visual search engine, as a follow up of the contribution I made to my startup. It is another application to expose knowledge graphs, following Freebase, Google’s knowledge graph, Wikidata, startups on visual knowledge graphs such as Blippar, Unigraph.rocks, XDiscovery (the startup I co-founded); mobile apps exposing knowledge graph (like the premium app I designed for iOS — Learn Discovery — Mindmap of Wikipedia).
Nifty.works is also a platform to edit and share knowledge maps.
So far it provide context for a subset of Wikidata, paired with English Wikipedia. It could be made for all other languages that are poorly addressed by research on NLP, semantic graphs and knowledge graphs (e.g. Thai, Chinese, Hispanic, Japanese and also Hispanic, French and German ). Each encyclopedia reveals in-depth insights of cultures, reflected by information written in local languages — as example, you will find much more information on French history in the French version. You got the idea.
You can access the knowledge graph with free API RESTful and built your own UI and UX — you don’t have to stick to visual graphs!
Here I visualise maps as a graphs.
Visualise, edit, share
Green circles highlight topics linked to each other.
To go for serendipity and visualise maps as trees, simply add spanningTreep parameter in the url — this is more responsive and you can explore maps of about 600 topics in full SVG resolution!
A shared map of 500+ nodes:
You can edit a map so to delete topics; add selected topics; clear out not-necessary topics. Once you’re done: share it!
As example, you may want to build maps that only show the amino acids and proteins they build. Or history of a place. Or bring people to discuss how religions are actually interconnected, and so the fabric of our cultures.
And if you really feel it, you can check out a prototype to gamify knowledge!
Despite not a full gamification here, nevertheless you can take the challenge to help Elon Musk colonise Mars, as fast as you can.
Once you’re done, you can share a map, call out a friend to expand it, share it with another person who expand it share it and see where it goes!
Tag with #chainLetterKnowledge and let’s see the journey that “math” have to do to reach out “music”; or “unconditional love” to reach out … well, who you love!
Which is your beloved person(s) background? Her(theirs) story or traditions?
What’s next? Sharing ideas, interests and opinions
I asked Stephen and Mahmoud if they would like to publish seealso.org — a site collecting visualisations built upon wikipedia — I received their precious feedback asking if this project is open source or not.
I haven’t released the code, and actually I am looking for opportunities, projects or teams to enjoy and bring value — so, busy on it!
I would LOVE to travel (at costs covered!) in development areas to share and teach kids and schools be autonomous in putting their hands on this or similar project, be autonomous in diving into their own learning things.
I also care about governance of innovation. So, I would be happy to experiment with licensing aiming to redistribute value that may be generated through open projects.
As example, wikipedia is free, but Google capitalised it to empower its search engine.
It is freely available for end-users, but I’d like to find solutions for commercial entities or wiki-based commercial projects (e.g. Wikiwand if they would be interested in!).
What’s Next ? Learning steps
‘Kkey this is a next call for me. I am learning on machine learning (introductory courses in Coursera) and if you work with patterns recognition, small-world networks, deep-learning, AI and bots — oh drop a line!
I can compute small-data knowledge graphs (e.g. a knowledge graph about neuroscience) and we can pipe it to a general graph.
The idea is aggregating knowledge graphs of different expertise, so to traverse it according to personal background.
I think that the quality of the knowledge graph I can provide as seminal point is good enough to be used as trainer model.
As example, the knowledge graph on Wikipedia has consistent results with Google Knowledge Graph research on vector paragraphs. I did other tests on other datasets, benchmarking them with state-of-the-art recommenders in industrial domains (tests Vs yummly.com, Thomson&Reuters, crunchbase).
I would like to contribute in research for automating production of new knowledge.
The impact I pursue is to help reducing asymmetries among diverse social environments and cultural backgrounds, and promote science as possibility to doubt over the hype of holding an elitarian truth.