OUseful.Info, the blog…

To corrupt a well known saying, “cook a man a meal and he’ll eat it; teach a man a recipe, and maybe he’ll cook for you…”, I thought it was probably about time I posted the recipe I’ve been using for laying out Twitter friends networks using Gephi, not least because I’ve been generating quite a few network files for folk lately, giving them copies, and then not having a tutorial to point them to. So here’s that tutorial…

The starting point is actually quite a long way down the “how did you that?” chain, but I have to start somewhere, and the middle’s easier than the beginning, so that’s where we’ll step in (I’ll give some clues as to how the beginning works at the end…;-)

Here’s what we’ll be working towards: a diagram that shows how the people on Twitter that @wiredUK follows follow each other:

The tool we’re going to use to layout this graph from a data file is a free, extensible, open source, cross platform Java based tool called Gephi. If you want to play along, download the datafile. (Or try with a network of your own, such as your Facebook network.)

From the Gephi file menu, Open the appropriate graph file:

Import the file as a Directed Graph:

The Graph window displays the graph in a raw form:

Sometimes a graph may contain nodes that are not connected to any other nodes. (For example, protected Twitter accounts do not publish – and are not published in – friends or followers lists publicly via the Twitter API.) Some layout algorithms may push unconnected nodes far away from the rest of the graph, which can affect generation of presentation views of the network, so we need to filter out these unconnected nodes. The easiest way of doing this is to filter the graph using the Giant Component filter.

To colour the graph, I often make us of the modularity statistic. This algorithm attempts to find clusters in the graph by identifying components that are highly interconnected.

This algorithm is a random one, so it’s often worth running it several times to see how many communities typically get identified.

A brief report is displayed after running the statistic:

While we have the Statistics panel open, we can take the opportunity to run another measure: the HITS algorithm. This generates the well known Authority and Hub values which we can use to size nodes in the graph.

The next step is to actually colour the graph. In the Partition panel, refresh the partition options list and then select Modularity Class.

Choose appropriate colours (right click on each colour panel to select an appropriate colour for each class – I often select pastel colours) and apply them to the graph.

The next thing we want to do is lay out the graph. The Layout panel contains several different layout algorithms that can be used to support the visual analysis of the structures inherent in the network; (try some of them – each works in a slightly different way; some are also better than others for coping with large networks). For a network this size and this densely connected,I’d typically start out with one of the force directed layouts, that positions nodes according to how tightly linked they are to each other.

When you select the layout type, you will notice there are several parameters you can play with. The default set is often a good place to start…

Run the layout tool and you should see the network start to lay itself out. Some algorithms require you to actually Stop the layout algorithm; others terminate themselves according to a stopping criterion, or because they are a “one-shot” application (such as the Expansion algorithm, which just scales the x and y values by a given factor).

We can zoom in and out on the layout of the graph using a mouse wheel (on my MacBook trackpad, I use a two finger slide up and down), or use the zoom slider from the “More options” tab:

To see which Twitter ID each node corresponds to, we can turn on the labels:

This view is very cluttered – the nodes are too close to each other to see what’s going on. The labels and the nodes are also all the same size, giving the same visual weight to each node and each label. One thing I like to do is resize the nodes relative to some property, and then scale the label size to be proportional to the node size.

Here’s how we can scale the node size and then set the text label size to be proportional to node size. In the Ranking panel, select the node size property, and the attribute you want to make the size proportional to. I’m going to use Authority, which is a network property that we calculated when we ran the HITS algorithm. Essentially, it’s a measure of how well linked to a node is.

The min size/max size slider lets us define the minimum and maximum node sizes. By default, a linear mapping from attribute value to size is used, but the spline option lets us use a non-linear mappings.

I’m going with the default linear mapping…

We can now scale the labels according to node size:

Note that you can continue to use the text size slider to scale the size of all the displayed labels together.

This diagram is now looking quite cluttered – to make it easier to read, it would be good if we could spread it out a bit. The Expansion layout algorithm can help us do this:

A couple of other layout algorithms that are often useful: the Transformation layout algorithm lets us scale the x and y axes independently (compared to the Expansion algorithm, which scales both axes by the same amount); and the Clockwise Rotate and Counter-Clockwise Rotate algorithm lets us rotate the whole layout (this can be useful if you want to rotate the graph so that it fits neatly into a landscape view.

The expanded layout is far easier to read, but some of the labels still overlap. The Label Adjust layout tool can jiggle the nodes so that they don’t overlap.

(Note that you can also move individual nodes by clicking on them and dragging them.)

So – nearly there… The final push is to generate a good quality output. We can do this from the preview window:

The preview window is where we can generate good quality SVG renderings of the graph. The node size, colour and scaled label sizes are determined in the original Overview area (the one we were working in), although additional customisations are possible in the Preview area.

To render our graph, I just want to make a couple of tweaks to the original Default preview settings: Show Labels and set the base font size.

Click on the Refresh button to render the graph:

Oops – I overdid the font size… let’s try again:

Okay – so that’s a good start. Now I find I often enter into a dance between the Preview ad Overview panels, tweaking the layout until I get something I’m satisfied with, or at least, that’s half-way readable.

How to read the graph is another matter of course, though by using colour, sizing and placement, we can hopefully draw out in a visual way some interesting properties of the network. The recipe described above, for example, results in a view of the network that shows:

- groups of people who are tightly connected to each other, as identified by the modularity statistic and consequently group colour; this often defines different sorts of interest groups. (My follower network shows distinct groups of people from the Open University, and JISC, the HE library and educational technology sectors, UK opendata and data journalist types, for example.)
- people who are well connected in the graph, as displayed by node and label size.

Here’s my final version of the @wiredUK “inner friends” network:

You can probably do better though…;-)

To recap, here’s the recipe again:

- filter on connected component (private accounts don’t disclose friend/follower detail to the api key i use) to give a connected graph;
- run the modularity statistic to identify clusters; sometimes I try several attempts
- colour by modularity class identified in previous step, often tweaking colours to use pastel tones
- I often use a force directed layout, then Expansion to spread to network out a bit if necessary; the Clockwise Rotate or Counter-Clockwise rotate will rotate the network view; I often try to get a landscape format; the Transformation layout lets you expand or contract the graph along a single axis, or both axes by different amounts.
- run HITS statistic and size nodes by authority
- size labels proportional to node size
- use label adjust and expand to to tweak the layout
- use preview with proportional labels to generate a nice output graph
- iterate previous two steps to a get a layout that is hopefully not completely unreadable…

Got that?!;-)

Finally, to the return beginning. The recipe I use to generate the data is as follows:

1. grab a list of twitter IDs (call it L); there are several ways of doing this, for example: obtain a list of tweets on a particular topic by searching for a particular hashtag, then grab the set of unique IDs of people using the hashtag; grab the IDs of the members of one or more Twitter lists; grab the IDs of people following or followed by a particular person; grab the IDs of people sending geo-located tweets in a particular area;
2. for each person P in L, add them as a node to a graph;
3. for each person P in L, get a list of people followed by the corresponding person, e.g. Fr(P)
4. for each X in e.g. Fr(P): if X in Fr(P) and X in L, create an edge [P,X] and add it to the graph
5. save the graph in a format that can be visualised in Gephi.

To make this recipe, I use Tweepy and a Python script to call the Twitter API and get the friends lists from there, but you could use the Google Social API to get the same data. There’s an example of calling that API using Javscript in my “live” Twitter friends visualisation script (Using Protovis to Visualise Connections Between People Tweeting a Particular Term) as well as in the A Bit of NewsJam MoJo – SocialGeo Twitter Map.

As the focus for this week’s episode [airs live Tues 21/6/11 at 19.32 UK time, or catch it via the podcast] in the OU co-produced season of programmes on openness with Click (radio), the BBC World Service radio programme formerly known as Digital Planet, we’re looking at one or two notions of diversity…

If you’re a follower of pop technology, over the last week or two you will probably have already come across Eli Pariser’s new book, The Filter Bubble: What The Internet Is Hiding From You, or his TED Talk on the subject:

Eli Pariser, :The Filter Bubble”, TED Talks

It could be argued that this is the Filter Bubble in action… how likely is it, for example, that a randomly selected person on the street would have heard of this book?

To support the programme, presenter Gareth Mitchell has been running an informal experiment on the programmes Facebook page: Help us with our web personalisation experiment!! The idea? To see what effect changing personalisation settings on Google has on a Google search for the word “Platform”. (You can see results of the experiment from Click listeners around the world on the Facebook group wall… Maybe you’d like to contribute too?)

It might surprise you to learn that Google results pages – even for the same search word – do not necessarily always give the same results, something I’ve jokingly referred to previously as “the end of Google Ground Truth”, but is there maybe a benefit to having very specifically focussed web searches (that is, very specific filter bubbles)? I think in certain circumstances there may well be…

Take education, or research, for example. Sometimes, we want to get the right answer to a particular question. In times gone by, we might have asked a librarian for help, if not to such a particular book or reference source, at least to help us find one that might be appropriate for our needs. Nowadays, it’s often easier to turn to a web search engine than it is to find a librarian, but there are risks in doing that: after all, no-one really knows what secret sauce is used in the Google search ranking algorithm that determines which results get placed where in response to a particular search request. The results we get may be diverse in the sense that they are ranked in part by the behaviour of millions of other search engine users, but from that diversity do we just get – noise?

As part of the web personalisation/search experiment, we found that for many people, the effects of changing personalisation settings had no noticeable effect on the first page of results returned for a search on the word “platform”. But for some people, there were differences… From my own experience of making dozens of technology (and Formula One!) related searches a day, the results I get back for those topics hen I’m logged in to Google are very different to when I have disabled the personalised reslults. As far as my job goes, I have a supercharged version of Google that is tuned to return particular sorts of results – code snippets, results from sources I trust, and so on. In certain respects, the filter bubble is akin to my own personal librarian. In this particular case, the filter bubble (I believe), works to my benefit.

Indeed, I’ve even wondered before whether a “trained” Google account might actually be a valuable commodity: Could Librarians Be Influential Friends? And Who Owns Your Search Persona?. Being able to be an effective searcher requires several skills, including the phrasing of the search query itself, the ability to skim results and look for signals that suggest a result is reliable, and the ability to refine queries. (For a quick – and free – mini-course on how to improve your searching, check out the OU Library’s Safari course.) But I think it will increasingly rely on personalisation features…which means you need to have some idea about how the personalisation works in order to make the most of its benefits and mitigate the risks.

To take a silly example: if Google search results are in part influenced by the links you or your friends share on Twitter, and you follow hundreds of spam accounts, you might rightly expect your Google results to be filled with spam (because your friends have recommended them, and you trust your friends, right? That’s one of the key principles of why social search is deemed to be attractive.)

As well as the content we discover through search engines, content discovered through social networks is becoming of increasing importance. Something I’ve been looking at for some time is the structure of social networks on Twitter, in part as a “self-reflection” tool to help us see where we might be situated in a professional social sense based on the people we follow and who follow us. Of course, this can sometimes lead to incestuous behaviour, where the only people talking about a subject are people who know each other.

For example, when I looked at the connection of people chatting on twitter about Adam Curtis’ All Watched Over By Machines of Loving Grace documentary, I was surpised to see it defined a large part of the UK’s “technology scene” that I am familiar with from my own echochamber…

#awobmolg echo chamber

So what do I mean by echochamber? In the case of Twitter, I take it to refer to a group of people chatting around a topic (as for example, identified by a hashtag) who are tightly connected in a social sense because they all follow one another anyway… (To see an example of this, for a previous OU/Click episode, I posted a simple application (it’s still there), to show the extent to which people who had recently used the #bbcClickRadio hashtag on Twitter were connected.)

As far as diversity goes, if you follow people who only follow each other, then it might be that the only ideas you come across are ideas that keep getting recycled by the same few people… Or it might be the case that a highly connected group of people shows a well defined special interest group on a particular topic….

To get a feel for what we can learn about our own filter bubbles in Twitterspace, I had a quick look at Gareth Mitchell’s context (@garethm on Twitter). One of the dangers of using public apps is that anyone can do this sort of analysis of course, but the ethics around my using Gareth as a guinea pig in this example is maybe the topic of another programme…!

So, to start with, let’s see how tightly connected Gareth’s Twitter friends are (that is, to what extent do the people Gareth follows on Twitter follow each other?):

The social graph showing how @garethm’s friends follow each other

The nodes represent people Gareth follows, and they have been organised into coloured groups based on a social network analysis measure that tries to identify groups of tightly interconnected individuals. The nodes are sized according to a metric known as “Authority”, which reflects the extent to which people are followed by other members of the network.

A crude first glance at the graph suggests a technology (purple) and science (fluorine-y yellowy green) cluster to me, but Gareth might be able to label those groups differently.

Something else I’ve started to explore is the extent to which other people might see us on Twitter. One way of doing this is to look at who follows you; another is to have a peek at what lists you’ve been included on, along with who else is on those lists. Here’s a snapshot of some of the lists (that actually have subscribers!) that Gareth is listed on:

The flowers are separate lists. People who are on several lists are caught on the spiderweb threads connecting the list flowers… In a sense, the lists are filter bubbles defined by other people into which Gareth has been placed. To the left in the image above, we see there are a few lists that appear to share quite a few members: convergent filters?!

In order to try to looking outside these filter bubbles, we can get an overview of the people that Gareth’s friends follow that Gareth doesn’t follow (these are the people Gareth is likely to encounter via retweets from his friends):

Who @garethm’s friends follow that @garethm doesn’t follow…

My original inspiration for this was to see whether or not this group of people would make sense as recommendations for who to follow, but if we look at the most highly followed people, we see this may not actually make sense (unless you want to follow celebrities!;-)

Popular friends of Gareth’s that he doesn’t follow…

By way of a passing observation, it’s also worth noting that the approach I have taken to constructing the “my friends friends who aren’t my friends” graph tends to place “me” at the centre of the universe, surrounded by folk who are a just a a friend of a friend away…

For extended interviews and additional material relating to the OU/Click series on openness, make sure you visit Click (#bbcClickRadio) on OpenLearn.

I’ve been tinkering with some of my “List Intelligence” code again, and thought it worth capturing some examples of the sort of network exploration recipes I’m messing around with at the moment.

Let’s take @jiscCetis as an example; this account follows no-one, is followed by a few, hasnlt much of a tweet history and is listed by a handful of others.

Here’s the follower network, based on how the followers of @jiscetis follow each other:

There are three (maybe four) clusters there, plus all the folk who don’t follow any of the @jisccetis’ followers…: do these follower clusters make any sort of sense I wonder? (How would we label them…?)

The next thing I thought to do was look at the people who were on the same lists as @jisccetis, and get an overview of the territory that @jisccetis inhabits by virtue of shared list membership.

Here’s a quick view over the folk on lists that @jisccetis is a member of. The nodes are users named on the lists that @jisccetis is named on, the edges are undirected and join indivduals who are on the same list.

Plotting “co-membership” edges is hugely expensive in terms of upping the edge count that has to be rendered, but we can use a directed bipartite graph to render the same information (and arguably even more information); here, there are two sorts of nodes: lists, and the memvers of lists. Edges go from members to listnames (I should swap this direction really to make more sense of authority/hub metrics…?)

Another thing I thought I’d explore is the structure of the co-list membership community. That is, for all the people on the lists that @jisccetis is a member of, how do those users follow each other?

It may be interesting to explore in a formal way the extent to which the community groups that appear to arise from the friending relationships are reflected (or not) by the make up of the lists?

It would probably also be worth trying to label the follower group – are there “meaningful” (to @jisccetis? to the @jisccetis community?) clusters in there? How would you label the different colour groupings? (Let me know in the comments…;-)

I didn’t follow the twitter backchannel surrounding the second part of Adam Curtis’ documentary series All Watched Over by Machines of Loving Grace, but I did see a tweet that made me grab a snapshot of the friend connections between folk who were tweeting around it using the awobmolg hashtag. The following shows the major connected component (unconnected tweeps are filtered out) of a network built up from folk using the hashtag and how they’re connected (node size a function of authority, I think?):

I guess you could argue that this provides a glimpse over part of the UK’s tweeting digerati community, though I’m not sure what part…? Nodes are layed out using a force directed layout algorithm and clustered according to modularity group. I think I can spot culture/education sector, open gov/data advocacy and media folk clusters in there. If you want to look deeper, here’s the data set I grabbed as a Gephi/gdf file.

@mhawksey also managed to grab approx 1500 or so of the #awobmolg tagged tweets into the Archive tab of this spreadsheet: awobmolg tweets

And by the by, here are cotags used in tweets alongside the #awobmolg tagged tweets I captured:

FIFA
EgyptsLostCities
Fuller
Cybernetics
BBC
therapeuticcommunity
Chaos
synergy
AWOBMOLG
Springwatch
TheMatrix
BGT
gameofthrones
politics
awobmolg
Bruntland
BBC2
anarchy
TheMotherofAllDemos
downsideofrain
TheStoryofIreland
marking
springwatch
RDLaing
powerofTV
spacearchaeology
gkchesterton
bbc
poetry
uep
notanetutopian
boomboom
systems
Bilderberg
adamcurtis
PaulMertonsHollywood
biofeedbackloop
allwatchedoverbymachinesoflovinggrace
socialmediacommodification
GameOfThrones
curtis
mature
bioweb
OctoberGallery
Freedom
shsn
BlatterOut
IIW
superinjunctions
banks
ple
psychoville
bgt
wtf
bollocks
AWObMoLG
ecosystems
buckyballs
Bbc2
alife
glee
anarchism
NIN
Observer
StarttheWeek
ecosystem
Twitter
silos
fb
UKUncut
Amortality
feedbackloop
honestly
eek
balancedequilibrium
Awobmolg
anarchist
bbc2
BBCTwo

ps via @josswinn, an archive of Adam Curtis films

Putting together a couple of tricks from recent posts (Visualising Vodafone Mclaren F1 Telemetry Data in Gephi and PDF Data Liberation: Formula One Press Release Timing Sheets), I thought I’d have a little play with the timing sheet data in Gephi…

The representations I have used to date are graph based, with each node corresponding a particular lap performance by a particular driver, and edges connecting consecutive laps.

**If you want to play along, you’ll need to download Gephi and this data file: F1 timing, Malaysia 2011 (NB it’s not throughly checked… glitches may have got through in the scraping process:-(**

The nodes carry the following data, as specified using the GDF format:

• name VARCHAR: the ID of each node, given as driverNumber_lapNumber (e.g. 12_43)
• label VARCHAR: the name of the driver (e.g. S. VETTEL
• driverID INT: the driver number (e.g. 7)
• driverNum VARCHAR: an ID for the driver of the lap (e.g. driver_12
• team VARCHAR: the team name (e.g. Vodafone McLaren Mercedes)
• lap INT: the lap number (e.g. 41)
• pos INT: the position at the end of the lap (e.g. 5)
• pitHistory INT: the number of pitstops to date (e.g. 2)
• pitStopThisLap DOUBLE: the duration of any pitstop this lap, else 0 (e.g. 12.321)
• laptime DOUBLE: the laptime, in seconds (e.g. 72.125)
• lapdelta DOUBLE: the difference between the current laptime and the previous laptime (e.g. 1.327)
• elapsedTime DOUBLE: the summed laptime to date (e.g. 1839.021)
• elapsedTimeHun DOUBLE: the elapsed time divided by a hundred (e.g. )

Using the geolayout with an equirectangular (presumably this means Cartesian?) layout, we can generate a range of charts simply by selecting suitable co-ordinate dimensions. For example, if we select the laptime as the y (“latitude”) co-ordinate and x (“longitude”) as the lap, filtering out the nodes with a null laptime value, we can generate a graph of the form:

We can then tweak this a little – e.g. colour the nodes by driver (using a Partition based coluring), and edges according to node, resize the nodes to show the number of pit stops to date, and then filter to compare just a couple of drivers :

This sort of lap time comparison is all very well, but it doesn’t necessarily tell us relative track positions. If we size the nodes non-linearly according to position, with a larger size for the “smaller” numerical position (so first is less than second, and hence first is sized larger than second), we can see whether the relative positions change (in this case, they don’t…)

Another sort of chart we might generate will be familiar to many race fans, with a tweak – simply plot position against lap, colour according to driver, and then size the nodes according to lap time:

Again, filtering is trivial:

If we plot the elapsed time against lap, we get a view of separations (deltas between cars are available in the media centre reports, but I haven’t used this data yet…):

In this example, lap time flows up the graph, elapsed time increases left to right. Nodes are coloured by driver, and sized according to postion. If a driver has a hight lap count and lower total elapsed time than a driver on the previous lap, then it’s lapped that car… Within a lap, we also see the separation of the various cars. (This difference should be the same as the deltas that are available via FIA press releases.)

If we zoom into a lap, we can better see the separation between cars. (Using the data I have, I’m hoping I haven’t introduced any systematic errors arising from essentially dead reckoning the deltas between cars…)

Also note that where lines between two laps cross, we have a change of position between laps.

[ADDED] Here’s another view, plotting elapsed time against itself to see where folk are on the track-as-laptime:

Okay, that’s enough from me for now.. Here’s something far more beautiful from @bencc/Ben Charlton that was built on top of the McLaren data…

First up, a 3D rendering of the lap data:

And then a rather nice lap-by-lap visualisation:

So come on F1 teams – give us some higher resolution data to play with and let’s see what we can really do… ;-)

PS I see that Joe Saward is a keen user of Lap charts…. That reminds me of an idea for an app I meant to do for race days that makes grabbing position data as cars complete a lap as simple as clicking…;-) Hmmm….

PPS for another take of visualising the timing data/timing stats, see Keith Collantine/F1Fanatic’s Malaysia summary post.

Last year, I popped up an occasional series of posts visualising captures of the telemetry data that was being streamed by the Vodoafone McLaren F1 team (F1 Data Junkie).

I’m not sure what I’m going to do with the data this year, but being a lazy sort, it struck me that I should be able to visualise the data using Gephi (using in particular the geo layout that lets you specify which node attributes should be used as x and y co-ordinates when placing the nodes.

Taking a race worth of data, and visualising each node as follows (size as throttle value, colour as brake) we get something like this:

(Note that the resolution of the data is 1Hz, which explains the gaps…)

It’s possible to filter the data to show only a lap’s worth:

We could also filter out the data to only show points where the throttle value is above a certain value, or the lateral acceleration (“G-force”) and so on… or a combination of things (points where throttle and brake are applied, for example). I’ll maybe post examples of these using data from this year’s races…. err..?;-)

For now though, here’s a little video tour of Gephi in action on the data:

What I’d like to be able to do is animate this so I could look at each lap in turn, or maybe even animate an onion skin of the “current” point and a couple of previous ones) but that’s a bit beyond me… (for now….?!;-) If you know how, maybe we should talk?!:-)

[Thanks to McLaren F1 for streaming this data. Data was captured from the McLaren F1 website in 2010. I believe the speed, throttle and brake data were sponsored by Vodafone.]

PS If McLaren would like to give me some slightly higher resolution data, maybe from an old car on a test circuit, I’ll see what I can do with it… Similarly, any other motor racing teams in any other formula who have data they’d like to share, I’m happy to have a play… I’m hoping to go to a few of the BTCC races this year, so I’d particularly like to hear from anyone from any of those teams, or teams in the supporting races:-) If a Ginetta Junior team is up for it, we might even be able to get an education/outreach thing going into school maths, science, design and engineering clubs…;-)

Following on from Social Networks on Delicious, here’s a view over my delicious network (that is, the folk I “follow” on delicious) and the dominant tags they use:

The image is created from a source file generated by:

1) grabbing the list of folk in my delicious network;
2) grabbing the tags each of them uses;
3) generating a bipartite network specification graph containing user and edge nodes, with weighted links corresponding to the number of times a user has used a particular tag (i.e. the number of bookmarks they have bookmarked using that tag).

Because the original graph is a large, sparse one (many users define lots of tags but only use them rarely), I filtered the output view to show only those tags that have been used more than 150 times each by any particular user, based on the weight of each edge (remember, the edge weight describes the number of times a used has used a particular tag). (So if every user had used the same tag up to but not more 149 times each, it wouldn’t be displayed). The tag nodes are sized according to the number of users who have used the tag 150 or more times.

I also had a go at colouring the nodes to identify tags used heavily by a single user, compared to tags heavily used by several members of my network.

Here’s the Python code:

import urllib, simplejson

def getDeliciousUserNetwork(user,network):
  url='http://feeds.delicious.com/v2/json/networkmembers/'+user
  data = simplejson.load(urllib.urlopen(url))
  for u in data:
    network.append(u['user'])
    #time also available: u['dt']
  #print network
  return network

def getDeliciousTagsByUser(user):
  tags={}
  url='http://feeds.delicious.com/v2/json/tags/'+user
  data = simplejson.load(urllib.urlopen(url))
  for tag in data:
    tags[tag]=data[tag]
  return tags

def printDeliciousTagsByNetwork(user,minVal=2):
  f=openTimestampedFile('delicious-socialNetwork','network-tags-' + user+'.gdf')
  f.write(gephiCoreGDFNodeHeader(typ='delicious')+'\n')
 
  network=[]
  network=getDeliciousUserNetwork(user,network)

  for user in network:
    f.write(user+','+user+',user\n')
  f.write('edgedef> user1 VARCHAR,user2 VARCHAR,weight DOUBLE\n')
  for user in network:
    tags={}
    tags=getDeliciousTagsByUser(user)
    for tag in tags:
      if tags[tag]>=minVal:
         f.write(user+',"'+tag.encode('ascii','ignore') + '",'+str(tags[tag])+'\n')
  f.close()

Looking at the network, it’s possible to see which members of my network are heavy users of a particular tag, and furthermore, which tags are heavily used by more than one member of my network. The question now is: to what extent might this information help me identify whether or not I am following people who are likely to turn up resources that are in my interest area, by virtue of the tags used by the members of my network.

Picking up on the previous post on Social Networks on Delicious, might it be worth looking at the tags used heavily by my followers to see what subject areas they are interested in, and potentially the topic area(s) in which they see me as acting as a resource investigator?

One of the many things that the delicious social networking site appears to have got wrong is how to gain traction from its social network. As well as the incidental social network that arises from two or more different users using the same tag or bookmarking the same resource (for example, Visualising Delicious Tag Communities Using Gephi), there is also an explicit social network constructed using an asymmetric model similar to that used by Twitter: specifically, you can follow me (become a “fan” of me) without my permission, and I can add you to my network (become a fan of you, again without your permission).

Realising that you are part of a social network on delicious is not really that obvious though, nor is the extent to which it is a network. So I thought I’d have a look at the structure of the social network that I can crystallise out around my delicious account, by:

1) grabbing the list of my “fans” on delicious;
2) grabbing the list of the fans of my fans on delicious and then plotting:
2a) connections between my fans and and their fans who are also my fans;
2b) all the fans of my fans.

(Writing “fans” feels a lot more ego-bollox than writing “followers”; is that maybe one of the nails in the delicious social SNAFU coffin?!)

Here’s the way my “fans” on delicious follow each other (maybe? I’m not sure if the fans call always grabs all the fans, or whether it pages the results?):

The network is plotted using Gephi, of course; nodes are coloured according to modularity clusters, the layout is derived from a Force Atlas layout).

Here’s the wider network – that is, showing fans of my fans:

In this case, nodes are sized according to betweenness centrality and coloured according to in-degree (that is, the number of my fans who have this people as fans). [This works in so far as we're trying to identify reputation networks. If we're looking for reach in terms of using folk as a resource discovery network, it would probably make more sense to look at the members of my network, and the networks of those folk...)

If you want to try to generate your own, here's the code:

import simplejson

def getDeliciousUserFans(user,fans):
  url='http://feeds.delicious.com/v2/json/networkfans/'+user
  #needs paging? or does this grab all the fans?
  data = simplejson.load(urllib.urlopen(url))
  for u in data:
    fans.append(u['user'])
    #time also available: u['dt']
  #print fans
  return fans

def getDeliciousFanNetwork(user):
  f=openTimestampedFile("fans-delicious","all-"+user+".gdf")
  f2=openTimestampedFile("fans-delicious","inner-"+user+".gdf")
  f.write(gephiCoreGDFNodeHeader(typ="min")+"\n")
  f.write("edgedef> user1 VARCHAR,user2 VARCHAR\n")
  f2.write(gephiCoreGDFNodeHeader(typ="min")+"\n")
  f2.write("edgedef> user1 VARCHAR,user2 VARCHAR\n")
  fans=[]
  fans=getDeliciousUserFans(user,fans)
  for fan in fans:
    time.sleep(1)
    fans2=[]
    print "Fetching data for fan "+fan
    fans2=getDeliciousUserFans(fan,fans2)
    for fan2 in fans2:
      f.write(fan+","+fan2+"\n")
      if fan2 in fans:
        f2.write(fan+","+fan2+"\n")
  f.close()
  f2.close()

Years ago, I used the Javascript Infovis Toolkit to put together a handful of data visualisations around the idea of the “social life of a URL” by looking up bookmarked URLs on delicious and then seeing who had bookmarked them and using what tags (delicious URL History – Hyperbolic Tree Visualisation, More Hyperbolic Tree Visualisations – delicious URL History: Users by Tag). Whilst playing with some Twitter hashtag network visualisations today, I wondered whether I could do something similar based around delicious bookmark tags, so here’s a first pass attempt…

As a matter of course, delicious publishes RSS and JSON feeds from tag pages, optionally containing up to 100 bookmarked entries. Each item in the response is a bookmarked URL, along with details of the single individual person who saved that particular bookmark and the tags they used.

That is, for a particular tag on delicious we can trivially get hold of the 100 most recent bookmarks saved with that tag and data on:

- who bookmarked it;
- what tags they used.

Here’s a little script in Python to grab the user and tag data for each lak11 bookmark and generate a Gephi gdf file to represent the bipartite graph that associates users with the tags they have used:

import simplejson
import urllib

def getDeliciousTagURL(tag,typ='json', num=100):
  #need to add a pager to get data when more than 1 page
  return "http://feeds.delicious.com/v2/json/tag/"+tag+"?count=100"

def getDeliciousTaggedURLDetailsFull(tag):
  durl=getDeliciousTagURL(tag)
  data = simplejson.load(urllib.urlopen(durl))
  userTags={}
  uniqTags=[]
  for i in data:
    url= i['u']
    user=i['a']
    tags=i['t']
    title=i['d']
    if user in userTags:
      for t in tags:
        if t not in uniqTags:
          uniqTags.append(t)
        if t not in userTags[user]:
          userTags[user].append(t)
    else:
      userTags[user]=[]
      for t in tags:
        userTags[user].append(t)
        if t not in uniqTags:
          uniqTags.append(t)
  
  f=open('bookmarks-delicious_'+tag+'.gdf')
  f.write('nodedef> name VARCHAR,label VARCHAR, type VARCHAR\n')
  for user in userTags:
    f.write(user+','+user+',user\n')
  for t in uniqTags:
    f.write(t+','+t+',tag\n')

  f.write('edgedef> user VARCHAR,tag VARCHAR\n')
  for user in userTags:
    for t in userTags[user]:
      f.write(user+','+t+'\n')
  f.close()

tag='lak11'
getDeliciousTaggedURLDetailsFull(tag)

[Note to self: this script needs updating to grab additional results pages?]

Here’s an example of the output, in this case using the tag for Jim Groom’s Digital Storytelling course: ds106. The nodes are coloured according to whether they represent a user or a tag, and sized according to degree, and the layout is based on a force atlas layout with a few tweaks to allow us to see labels clearly.

Note that the actual URLs that are bookmarked are not represented in any way in this visualisation. The netwroks shows the connections between users and the tags they have used irrespective of whether the tags were applies to the same or different URLs. Even if two users share common tags, they may not share any common bookmarks…

Here’s another example, this time using the lak11 tag:

Looking at these networks, a couple of things struck me:

- the commonly used tags might express a category or conceptual tag that describes the original tag used to source the data;

- folk sharing similar tags may share similar interests.

Here’s a view over part of the LAK11 network with the LAK11 tag filtered out, and the Gephi ego filter applied with depth two to a particular user, in this case delicious user rosemary20:

The filtered view shows us:

- the tags a particular user (in this case, rosemary20) has used;

- the people who have used the same tags as rosemary20; note that this does not necessarily mean that they bookmarked any of the same URLs, nor that they are using the tags to mean the same thing*…

(* delicious does allow users to provide a description of a tag, though I’m not sure if this information is generally available via a public API?)

By sizing the nodes according to degree in this subnetwork, we can readily identify the tags commonly used alongside the tag used to source the data, and also the users who have used the largest number of identical tags.

PS it struck me that a single page web app should be quite easy to put together to achieve something similar to the above visualisations. The JSON feed from delicious is easy enough to pull in to any page, and the Protovis code library has a force directed layout package that works on a simple graph representation not totally dissimilar to the Gephi/GDF format.

If I get an hour I’ll try to have a play to put a demo together. If you beat me to it, please post a link to your demo (or even fully blown app!) in the comments:-)

Following on from Sketching the Structure of the UK Political Media Twittersphere, I did a quick trawl of US Senators and Congress members based on the lists maintained by @HuffPostPol, and visualised the connections between them in Gephi using the Force Layout algorithm:

Node size is proportional to betweenness centrality measured over the directed graph where edges go from one person to a person they follow. The edge set is limited to edges between members of the four @huffPostPol lists: republican-senators, democratic-senators, democratic-members, republican-members.