More science mapping tools - Local citation network tool and Open Knowledge Maps with concept graphs
Earlier this year, I blogged about how new citation indexes were impacting Science mapping tools such as Citespace, VOSviewer , Citation Gecko and more.
In this blog post, I will briefly talk about two more new mapping/visualization tools that caught my eye.
The first of them is Open Knowledge maps which has been around for a while, but has a beta that now supports concept graphs in partnership with Know-Center.
Open Knowledge Maps - tie up with Know Center's Concept Graph
The second is a open source tool called Location Citation Network by Tim Wölfle with uses either Microsoft Academic or Crossref as sources to create visualization of citation maps - somewhat similar to Citation Gecko and VOSviewer.
Location Citation Network by Tim Wölfle
Open Knowledge Map
Open Knowledge Map is a non-profit organization that supports open science and open infrastructure. The tool itself allows you to do a search of Pubmed or BASE (Bielefeld Academic Search), and visualizes the results in clusters.
In the example below, I searched for the string word stress in Pubmed. the tool then takes the top 100 papers and visualizes them into clusters based on text similarity.
Visualization of results from the search "Work Stress" from Pubmed in Open Knowledge Graph
The FAQ explains that it uses metadata (not full text) for clustering including "titles, abstracts, authors, journals, and subject keywords to create a word co-occurrence matrix between articles. On top of this matrix, we perform clustering and ordination algorithms. The labels for the sub-areas (bubbles) are generated from the subject keywords of the articles in this area. In cases where they are missing from the metadata, we approximate them from abstract and title"
Clusters that are closer are more similar in subject (though papers can only be assigned to 1 cluster), positions of papers in each cluster has no meaning.
On the right panel it shows the top 100 papers including links for open access papers.
This tool is currently in beta, so it's features are still limited. In the video below, they set out their vision which among other things allow users to manually edit and collaborate with others on such open knowledge maps (which are all made available via CC0).
Vision for collaborative editing in Open Knowledge Maps from Open Knowledge Maps on Vimeo.
I must admit, I have not been that impressed by this tool , until the recent update that introduces concept graph.
Concept graph
What is a "concept graph"? It's perhaps easier to show one.
A sample concept graph I generated
As far as I can tell it is a network graph of not just papers/publications as nodes (yellow) but also authors (red), keywords (blue) and publication year (lighter blue).
The help file tells me the size of the node indicates it's importance.
All in all this reminds me of the concept of a PID Graph mooted by Data Cite/Freya. With idea of linking various nodes representing research entities (e.g. publications, datasets, authors, funders etc) by Permanent IDs.
From Datacite blog post - introducing the PID Graph
How can you get to a concept graph from Open Knowledge map?
I see two ways to get to a concept graph from a open knowledge map.
Once you have generated a open knowledge map, you can either scroll down to the bottom of the page and click on "Try out concept graph" button or you can click on "Explore connections of this document" for any article listed on the right pane of any generated open knowledge map.
Click on "Explore connections of this document" below each article to generate concept maps
In the first case, it will generate a bare concept graph with the top 10 publication nodes (un-expanded) in the open knowledge graph showing while for the later case it will just put one publication node on the graph.
Below is an example of the earlier case..
10 publication nodes (unexpanded) appearing in concept map
I suspect many people upon seeing this will try to click/mouse over the yellow publication nodes and they will see this.
Mouse over publication nodes, click on the Green Plus to expand
The image above tells you that for this publication note, it is linked to a single year node, five author nodes and three keyword nodes.
But what do the numbers like (0/3) mean? This means currently 0 of the nodes are shown while 3 are available.
If you click on the big green plus button it will expand all the nodes around that publication.
Expanded nodes for concept map
The temptation now is to just get expanding nodes , but it will get very messy very quickly.
You can do things to simplify matters by hiding labels (appearing only in mouse overs). using the filters to hide nodes types you are not interested in, by years or even text in labels
Sample concept map with all author nodes filtered + labels turned off
Getting around in the concept graph
Upon generating the concept map, the starting help screen that popups, informs you that you can zoom in and out with the mouse wheel, delete nodes by selecting them and clicking on the trash button at the bottom right, drag around nodes etc.
As you hover over each node it shows all the connections in red lines to nodes currently expanded.
If you are like me, you will start expanding nodes on a whim and soon you get a complicated mess.
A complicated mess
The idea here is using a combination of methods, you can simplify things (e.g. by expanding only some nodes) enough to gain insight on how different nodes are connected.
With careful expansion of nodes hopefully you can explore and produce concept maps that help provide some insight.
Controlled expansion of concept graph
For example, the above graph shows that the keyword "Occupational stress factors" is used for 4 years (see the connections to the blue icons for 2016,2017,2018,2019), and is also co-used with 3 keywords. The keyword "Occupational stress factors" is actually connected to 13 documents but only 5 (denoted by yellow nodes) is shown as indicated by Documents(5/13). Similarly it is showing currently only 2 our of 8 authors nodes (in red) linked to this keyword.
Using advanced mode - Ring menu and subgraph
Clearly, to make the best use of the concept graph, you have to be able to selectively open nodes. This is where the advanced modes come in.
If you open enough nodes, it will eventually offer you the advanced mode which offers two functionalities, the Ring menu and the Subgraph.
The Ring Menu allows you to selectively open nodes.
The Ring Menu in the concept graph - right click and hold any node to generate
Right click and hold on any node in the concept graph and it will give you a ring menu which has 3 layers of ring around the original node (in this case a yellow document node).
The first ring around the original node, represents the number of nodes that are one hop away. The ring after that , nodes that are two hops away and the ring after that . nodes that are 3 hops away.
When you mouse over each segment of the ring menu it tells you the number available.
So in the example above, 3 hops away from the originally selected document node there are 206 author nodes but only 2 are visible/expanded.
In the ring menu, if you left click the segment representing it, up to 5 nodes will expand. For example if you click on the red ring on the outer most part of the ring menu , it will expand 5 author nodes that are 3 hops away. Click it again and it will expand 5 more.
If you wish more fine grained control of the number of nodes, right click and hold on the segment and it will give you a slider to control exactly how many nodes you want to be displayed.
Right click and hold on segments in the Ring Menu for a slider to appear
Though this feature is not very intuitive in my opinion it does give you very fine grained control of what to expand, though I expect normally, most people would expand the inner-most links to look at direct connections.
The other advanced filter is called Subgraphs which in a way is the inverse of ring menus. While ring menus are ways to selectively expand, Subgraphs allow you to merge/group expanded nodes together.
To do this, hold cltrl and use your mouse to select an area with the nodes you want to merge. Click on the merge button on the top left part of the filter panel.
The merger button appears once you have selected a few nodes in the concept map
This is what you will get.
Sample Subgraph in concept map
Hovering over sections will get you more information. e.g. Documents(8) are connected to Keywords(2).
Currently, I'm still wondering how best to use this feature. I would guess this is done for grouping up a lot of nodes that you all can treat as "one entity", greatly simplifying the concept graph. allowing you to focus on what parts you can.
Thoughts about the concept map feature
Overall, this concept map is a intriguing idea, the fact that you can drag nodes where-ever you want, trash nodes that are not useful makes this one step above Open Knowledge Maps itself (as it does not provide such customization features).
The main problem is that the advanced features are not intuitive at all, and for a typical researcher even if they figure out the mechanics they are still going to have problems figuring out how to use this tool.
It would also be nice to understand what type of relationships between node types is available. For example is it possible for a publication node to be connected directly to another? How about an author node to a keyword? What does it mean if a keyword node is connected to another keyword node?
This is of course a very new tool and you can help them out to improve the tool by doing the questionnaire (link on top left).
Local citation network
Local citation network is one of a growing number of tools that depend on Microsoft Academic and Crossref for functionality.
The idea is simple, either enter a list of dois in a text file for "input dois" or just enter one doi.
In the later case, the system will look up Microsoft Academic/Crossref for references made in that doi and use them as "input dois". In other words, you will be analysing relationships between references in the original source doi (I don't think the original doi is included?).
Two ways to input dois into Local Citation Network tool
It will then try to query the references in the input dois and query either Microsoft academic or Crossref to find the top 10 references with highest number of cites in "the local network" that aren't already input dois.
These 10 new references becomes the suggested papers and they are visualized as a star icon in the map.
In the image below, I have generated a citation map using 10.7717/peerj.4375 as the source doi.
Local Citation Network generated from 10.7717/peerj.4375
Citation network generated by Local Citation Network tool
In many ways the citation network generated is pretty standard if you have used tools like VOSviewers.
Each node represents a paper. Size of node represents how well cited the article is. Unlike many other citation visualizations where the vertical dimension doesn't denote anything (e.g. VOSviewer's), in this case older articles (by publication year) are placed towards the bottom and the newer ones are on top.
The edges or connections are also directed, pointing from citing paper to cited papers, as such in the visualizations all edges/connection are either pointing downwards (towards older paper) or horizontally (papers in the same publication year)
There is also a co-author network but I seldom find them useful.
Co-author network in Local citation network
Calculating "estimated completeness"
By default this tool uses Microsoft Academic Graph (you can enter your own key or use the common one) , but you can switch to Crossref.
Either way, this tool calculates a metric called "estimated completeness".
This allows you to tell how complete the local citation network generated is.
In the case of Microsoft Academic, it seems tells you the % of input articles that have reference lists.
Estimated Completeness of 27/32 = 84% of Local Citation Network
As noted in the help file, Crossref data allows a more sophisticated calculation of completeness.
Estimated Completeness calculation for citation network based on Crossref
Thoughts about Local Citation Network
It's pretty natural to compare this tool with Citation Gecko.
The main thing I like about this tool is you can just upload a text file of dois (I find I can do up to 500 dois). Doing so in Citation Gecko is a bit more indirect as you need to upload via Reference Manager or search.
It's also nice that you can upload just one source doi, to generate a citation map, while Citation Gecko requires you to upload a list of "Seed dois" (equalvant to input dois).
On the other hand, Local Citation Network tool has quite a few disadvantages
Firstly, as noted in the help file, because of the way it works it can only find older seminar papers as suggested papers but not later papers. Citation Gecko on the other hand offers two modes - cited by and citing modes.
Secondly, the interface for Local citation network can be improved, for example, the left pane of article makes up 50% of the screen and can't be reduced, so the visualization becomes squeezed
Lastly, the tool currently is fixed at 10 suggested articles while with Citation Gecko you can go beyond that and more importantly expand the network by setting suggested articles as seed/input articles to further find new articles. It is this iterative expansion that I think is important for exploration based tools.
That said, why not just use Citation Gecko? The unfortunate matter is as of now Citation Gecko now only uses Crossref which is fairly limited compared to Microsoft academic data which Local citation network tool uses.
In any case, both tools are open source and available on Github......
Conclusion
Most users are familiar with the current paradigm where you enter some keywords and you get back a list of articles. While such tools are intuitive to use, it is perhaps an area where we are approaching diminishing returns in terms of what can be done.
Exploration based tools such as Yewno that rely on visualization to help users explore and orient themselves are however a whole new ball game and developers are slowly feeling their way through, even as Scholarly data is becoming increasingly available for use.
Creating a interface that is easy to use and provide helpful support to users who often are not familiar with such tools seems to be the challenge!