Gephi is great for exploratory work, but I’ve seen it lure people into using methods that they don’t fully understand. The same problem that many statistics apps face.
IMO, a better approach to proper network analysis is to use a library such as the excellent igraph in R or Python together with a clear understanding of the measures.
There is a wonderful book called „network analysis literacy“ by Katharina Zweig which really helps with the latter.
Every once in a while I have a network related data science type problem to muddle through and I invariably want to try visualise some aspect of it, so I invariably try use Gephi for it, and it invariably leads to a frustrating experience for anything non trivial, and so I invariably end up doing it in code + graphviz instead.
To be fair, a large part of the problem is that many graph-related algorithms are simply very expensive (computationally). For example, many layouts (especially classical ones such as Fruchterman-Rheingold) are very slow. That often makes visualization frustrating when you have more than a couple of hundred nodes and/or a dense network. And we should acknowledge that the Gephi people have put a lot of work into making it work.
So network analysis evidently benefits from think-then-do approaches, while exploratory work is really hard.
I use Gephi a lot with work - it's helped explain lots of network complexity to leadership teams. They don't typically understand when you talk about risks of a network partition or node failure, but you visualize it and suddenly it's much more apparent. The circle layout, for me, works best with larger graphs.
My only quibble would be a better graph search function and highlighting.
I love it because it handled huge dependency graphs from Symbian and Yocto builds which showed the brutal level of interconnections and where the really big dependency magnets were (GUI toolkits like Qt depend on almost everything). Nothing else that I could find really showed that. It was a bit tricky to use sometimes but revolutionary in terms of giving me a gut understanding of the problems I was facing.
I’m a huge fan of Cytoscape.js. Not sure if it would be a competitor to Gephi as it’s just a JavaScript library but it’s very useful for things one might use D3 for. Not too not does it have the ability to draw, style, and animate the networks it has all the graph algorithms to do the analysis and traversal.
Is this once again another self contained one with plug-in architecture?
Or can it call from external like lisp-stat (sorry if not really apologetic) …
And given it is seem step into my current search for social simulation env, how can this visionisation give itself to statistical analysis result and then move into some parameters (programming) so we can simulate, collect, trial some strategy (both manual or limited old style programming like lisp or even go for new era AI which look at output pattern/input/policy …
Or more side track a bit but perhaps also important run or at least control under exploratory Jupiter-lab notebook. If not programming at least as documentation and testing / demo to fellow researcher or just student.
It definitely fulfilled my expectations for the joke part :))
Regardless, there is quite a bit of source available in the git repo so it shows it can be done even though it chugs quite a bit.
Our visual graph AI tool includes a GPU-accelerated take on gephi's flows and puts on the web, including a free GPU-accelerated tier with no-code UIs, embedding & control APIs (python, js, react, arrow), and deep pydata integration (Jupyter, RAPIDS, dashboarding like databricks & streamlit, ...): www.graphistry.com/get-started .
It's used a lot by folks doing fraud, IT, social, security, supply chains, anti-misinfo, finance, bio, etc. Mostly data scientists today, and as we have been launching no-code & low-code features, a diverse broader analyst community has been growing, who has been inspiring.
Gephi got a bit frozen in time due to the usual problem of struggling for post-phd sustainability by not building it in: I'm a big fan of the founders and their work, and just like Graphviz (att research canceled it), it was painful watching them having to leave something so cool. We prioritized sustainability as an engine for reliable & growable OSS, which has worked (ex: you may have heard of Apache Arrow, which we helped kick off). So our free SaaS tier aims to include everything in Gephi, and a lot missing in it for modern use: GPU accel, DB connectors & visual playbooks (already in self-hosted), visual graph ETL, and launching a bunch of graph AI stuff (entity linking, event scoring, recommendations, ... by automating UMAP and graph neural network flows). Likewise, we have + are steadily launching things not in Gephi yet you'd expect of modern team+enterprise tools like sharing, RBAC, SSO, daily-scanned docker/k8s/AMIs, etc. We are aiming for a model basically somewhere between gitlab and GitHub, and as we hit more sustainability, keep biasing for more free & OSS.
The good news is, years later.. it worked! We have reached sustainable growth & measurably best-in-class performance, so we are now growing, releasing more (including another big OSS visual auto-AI release just this week), and overall moving to next phases. If you like webgl, JS, or sales engineering (same industries you'd see in graph DBs), we are hiring for multiple roles in visual graph AI, and I'd very much love to chat :)
IMO, a better approach to proper network analysis is to use a library such as the excellent igraph in R or Python together with a clear understanding of the measures.
There is a wonderful book called „network analysis literacy“ by Katharina Zweig which really helps with the latter.