Google has launched a new, highly relevant capability. Google search now supports the Bacon Number for actors, based on building a small world social network, probably from spidering IMDb or the like. If you type in, for example, ‘bacon number laurence olivier’ you will see this:
September 14, 2012
Padraig Mac Carron and Ralph Kenna apply social network analysis to old texts — like The Iliad and Beowolf — to see if the relationships between the characters is like real social snetwork. If they are, that lends support for the premise that these tales are about real people, not just some shadows cast on the wall.
Padraig Mac Carron and Ralph Kenna, The Social Networks of Myths
Social networks have been widely studied in recent years; researchers have looked at the interconnectedness of groups like actors, musicians and co-authors of scientific texts. These networks share similar properties: they are highly connected, small worlds. They are assortative, which means that people tend to associate with people like themselves. And their degree distributions are usually scale-free — a small number of people tend to have lots of friends.
The myth networks were found to have some of the characteristics, including the small-world property and structural balance (related to the idea that the enemy of my enemy is my friend), typical of real-world networks.
Intentionally fictional narratives like “Harry Potter” also have these properties. However, “The Iliad” is assortative as well — a potential real-life indicator that these fictional networks lack. “Beowulf” is also assortative, but only if the main character, who is very different from the rest, is removed from the network. The “Tain,” [an Irish epic, which many believe to be entirely fictional] like the fictional networks we studied, is disassortative.
These and other features may corroborate scholars’ belief in the narratives’ historical basis: i.e., the societies underlying “The Iliad” and “Beowulf” may have traces of reality, while that of the “Tain” appears more artificial.
But just how fictional is the “Tain”? We looked further into the degree distributions of the social networks — at the frequency of popularity among characters. Like real networks, all three were scale-free, unlike any of the intentionally fictional narratives examined.
In this regard, the majority of the “Tain” and “Beowulf” were similarly realistic — except that in the Irish myth, the top six characters are all unrealistically well connected, giving it both fictional and real characteristics.
But there are 398 other characters in the “Tain,” and when we remove the weakest links (or single, direct encounters) between these characters and the Top 6, the narrative becomes as realistic as “Beowulf” from a social-network view. Perhaps these characters are amalgams of a number of entities that were fused as the narrative was passed down orally.
Our initial results may therefore corroborate existing interpretations of “Beowulf” and “The Iliad” as being at least partly historical. They also signal that the society in the “Tain” may have a similar level of historicity.
So, the next time you watch Troy or Beowulf, or some other movie based on old, old tales, remember that those were real people way beck when, that had friends and enemies, just like we do today.
- Is Mythology Like Facebook? (news.sciencemag.org)
- Physicists study the classics for hidden truths (esciencenews.com)
Imagine a not-too-distant future where machinery — servers, refrigerators, nuclear reactors — are on line, and following each other. They could respond to messages in programmatic ways, and we could use social media analysis tools to discover the ‘sentiment’ of subways and elevators.
Bruce Sterling called these messaging machines ‘spimes’ (although his concept went a bit further), but it’s clear we are rapidly moving into uncharted territory when machinery may independently who to follow and to talk with:
Social networks serve people as ways to communicate the way we live and work. A machine’s social network can serve similar purposes. The machines can have friends or even families that live in “clusters.” Each machine can learn from the individuals or communities in the collective group. They know when one is sick. They can relate to other machines and the way they feel.
At VMworld,in Monday’s keynote, the attendees saw a demo for how this might work. It shows how a social network populated with machines can spread word to each other. When one host finds an issue, it updates its activity stream. Other hosts and virtual machines will “like,” the update if they are having similar issues.
The VMware example points to an inevitable future. The machines will have a voice. They will communicate in increasingly human-like ways. In the near term, the advancements in the use of social technologies will provide contextual ways to manage data centers. Activity streams serve as the language that people understand. They help translate the interactions between machines so problems can be diagnosed faster.
By treating machines as individuals we can better provide visualizations to orchestrate complex provisioning and management tasks. That is inevitable in a world which requires more simple ways to orchestrate the increasingly dynamic nature for the ways we humans live and work with the machines among us.
Williams starts to creep me out when he — more or less — ascribes feelings to machines. But there is no doubt that machines could message each other, and act on those messages. And of course, they could listen to our messages, and act on them, such as the obvious scenarios of telling our thermostats to turn on the heat. But the not-so-obvious listening behaviors of future machines might be more interesting, like soda machines deciding what soda to order based on analyzing the demographic evidence of occupants in an office building.
That’s a bit more out there than sending the machine a tweet, asking for Izze Grapefruit soda instead of Fanta, and then getting a tweet back two days later from the soda machine — @419ParkAveSSoda1 — telling me that Izze is available in the machine. I notice I am now being followed by 37 soda machines in my neighborhood, and three of them favorited my tweet asking for Izze.
As hurricane Isaac batters New Orleans, seven years after Katrina, it’s worthwhile to ask a few questions about recovery after disasters, specifically why some places never seem to bounce back from disaster. And New Orleans has never really bounced back from Katrina.
New Orleans has the highest per capita murder rate in the US, and has led the country since the 1990s. In 1985, New Orleans had 421 murders, 85.4 per 100,000, a rate than has never been matched elsewhere, and which is more than 10 times the national average.
These are not unrelated facts: they are two sides of the same coin. New Orleans is in the region of the US with the lowest levels of social capital. Social capital is that cultural glue that ties people together, or, in its absence, leaves people as disconnected individuals, without links to each other and without common cause. Pierre Bourdieu defines social capital as
the sum of the resources, actual or virtual, that accrue to an individual or a group by virtue of possessing a durable network of more or less institutionalized relationships of mutual acquaintance and recognition.
And why does New Orleans seem to have such low social capital? Robert Putnam wrote about this issue in his masterwork, Bowling Alone:
Differences among the states on the underlying measures [of social capital] are substantial, with ratios of roughly three to one between high- and low-ranking states. Social trust, for example, ranges from 16 percent in Mississippi to 67 percent in North Dakota. The average number of associational memberships per capita varies from 1.3 in Louisiana and North Carolina to 3.3 in North Dakota. […] Even a cursory glance at America’s social capital resources leads one to ask, “Where in the world did these differences come from from?” Answering that question in detail is a task for another day, but this pattern has deep historical roots. Alexis de Tocqueville, patron saint of contemporary social capitalists, observed precisely the same patterns in his travels in the America of the 1030s, attributing it, at least in part, to patterns of settlement:
As one goes farther south [from New England], one finds a less active municipal life; the township has fewer officials, rights, and duties; the population does not exercise such a direct influence on affairs; the town meetings are less frequent and deal with fewer matters. For this reason the power of the elected official is comparatively greater and that of the voter less; municipal spirit is less awake and less strong…. Most of the immigrants who founded the northwestern states came from New England, and they brought the administrative habits of their old home to the new.
Well-trod paths of migration helped establish regional and local patterns of social capital in contemporary America.
Still more striking is the spatial correlation between low social capital at the end of the twentieth century and slavery in the first half of the nineteeth century. The more virulent the system of slavery then, the less civic the state today. Slavery was, in fact, a social system designed to destroy social capital among slaves and between slaves and freemen. Well-established networks of reciprocity among the oppressed would have raised the risk of rebellion, and egalitarian bonds of sympathy between slave and free would have undermined the very legitimacy of the system. After emancipation the dominant classes in the South continued to have a strong interest in inhibiting horizontal social networks. It is not happenstance that the lowest levels of community-based social capital are found where a century of plantation slavery was followed by a century of Jim Crow politics. Inequality and social solidarity are deeply incompatible.
So, as I read about the flooding this morning in Tammany Parrish on the outskirts of New Orleans, I thought about the effects of hundreds of years of disrupted social networks, of institutionalized inequality, and the barriers that we erect to stop people from connecting.
New Orleans won’t bounce back, and not because Katrina, Isaac, and other storms are so damaging. New Orleans may linger on as a reminder of a past age, but not one that we should mourn.
Megan Garber at The Atlantic caught my attention with a post about Twitter and the use of swearing:
We know, at this point, how the nation tweets. But what about how the nation swears?
The Ukrainian-based web development firm Vertaline, aiming to answer that question, scanned tweets posted from across 462 specific locations in the U.S. The team then isolated particular phrases from those tweets — one of those phrases being, yep, “fuck you,” which they tracked between July 14 and July 24, 2012. They then created a dynamic heatmap that portrays the density of the F-bomb-laden tweets as they were distributed geographically throughout each day of their date range, measured once per hour.
The screencapped maps above and below track our great nation’s keyword-ed conversation as it played out over a single day: Saturday, July 14. Particularly on the coasts, it seems — and even more particularly in L.A. and NYC – many, many fucks were given.
July 14 5:20
The company behind this, Vertaline, provides very minimal controls, but one thing is certain: people in LA and Buffalo say ‘fuck you’ a lot. Must be the song.
- Where Do Twitter F-Bombs Come From? Heat Map Shows Rudest Places (mashable.com)
- Where is the most cuss-friendly place in America? (digitaltrends.com)
- How America Swears: Here’s a Heatmap Tracking Twitter Profanity #TheAtlantic (drhiphop85.com)
August 14, 2012
New research suggests that people with suicidal tendencies don’t differ from others just because they think about suicidal acts, but because they are not embedded in social ‘neighborhoods’ where their online contacts know each other. It appears that they are isolated in their social graphs:
Spotting Suicidal Tendencies on Social Networks via MIT Technology Review
It turns out the people prone to suicide ideation have about the same number of friends as the control group. This alone does not seem to be a defining characteristic in the online world, where ‘friends’ are easy to come by. Neither does age or gender seem to be an identifying chaacteristic, which flies in the face of previous research.
The warning signals are more subtle, say Masuda and co. For example, people prone to suicide ideation are likely to be members of more community groups than the control group. That may be the result of spending longer online and of a desire to want to interact.
But a key indicator seems to be that these people are much less likely to be members of friendship triangles. In other words, they have fewer friends who also friends of each other. This low density of friendship triangles appears to be a crucial.
So, in terms out outreach, social network analysis — both about regarding content and network context — might be used in the future to help steer people thinking of suicide to help… or maybe we should just introduce their unacquainted friends to each other?
Ok “Ostracism is Community” sounds very Orwellian. It could very well be a mantra for a dictatorship based on isolating people or one of the lines from The Sphinx in Mystery Men.
But there is something to this mantra. The story starts with Stowe, as many stories do. In our discussions he had brought up mathematical based community identification algorithms. Now this is a question that has been plaguing graph theory specialists for a long time it turns out, and not something an astrophysicist could solve on the back of an envelope (bear in mind that in astronomy an order of magnitude is often seen as close enough).
So Stowe sent me a paper that proposed a new algorithm for the detection of communities in networks. Modularity is the key metric in this business and this paper is no exception. The quick explanation of modularity is that it is a measure of how a subnetwork is more tightlly knit than what randonmenss would indicate for the same group of nodes in that subnetwork. I will spare you the details of the paper as I my tenuous grasp of graph theory barely enabled me understand them myself, but suffice to say that small communities are very hard to delimit and that the calculation time required to perform this analysis is prohibitive for large networks.
What makes algorithmic community detection so difficult is a problem inherent in most algorithms: When do you make it stop? Mathematically, communities exist within networks, and in society the same is true. In todays nomenclature we often confuse networks with community. Where does the community end if it doesn’t take up the whole network? What is the maximum number of communities do we want to set in the network as the boundary condition?
The delimitation of communities is then a fundamental distinguishing factor between a community and a network, in fact it is a necessary condition of community. Communities are more closely knit, (as modularity calculations demonstrate) but additionally at least in real human communities, they also include goals or a a set of guiding principles, and the human boundary condition to community is a process of ex-communication or ostracism. You may ask: “How can something as vile and hurtful as ostracism be a part of something as inclusive and nurturing as community”?
In fact it is the ostracism that is making the community nurturing, because without ostracism the bounds of trust that are needed in communities are broken by the possible disruptive elements that may operate within that community. The community’s only answer to this is to expel the elements which break either the guiding principles of the community or injure the bonds of trust that enable it to take action.
The very word ostracism comes from an actual community that regurlarly kicked people out. The first recoded democracy in history in fact, Athens. Every year Athenians had a secret vote on one citizen they would chose to expel. They wrote the names of people they wanted to exclude on pieces of broken pottery called ostricon.
The Athenian who received the most “votes” was then kicked out of Athens or ostracized. For instance this Ostricon has the name of the Athenian statesman Cimon written on it. It’s kinda like voter recall but for everyone.
It sucks when the cool kids don’t want to hang out with you, I know (Astrophysicists are rarely part of the cool kid crowd in high school). Ostracism hurts when it happens, but everyone amongst us has told a troll to leave the discussion on a forum or thread, (Mr. Troll you know who you are). We have all stopped inviting the party guest who has to be right all night, diverges every discussion, starts fights and pisses every other guest off. We have excluded the carnivore at the vegetarian pot luck, the swing dancer at a tango night, or my favorite, the speed metal bassist in a country band. We have all done it, because we all cherish our communities and the goals they aim to achieve.
A recent startup called SNTMNT has opened its Trading Indicator API, which provides price predictions about the price of S&P 500 stocks. The company says its accuracy is about 64%, which suggests that users of the API will be consolidating these indicators with other sources of information.
The approach is a very detailed level of analysis of Twitter sentiment about the stocks, based on so-called ‘stock tickers’ — references to companies like AT&T, whose ticker is T. Many Twitter users talk about stocks, referring to the companies by ticker, which is preceded in most cases by a dollar sign, as in $T.
Johan Bollan and other researchers published a paper, Twitter mood predicts the stock market, back in 2010, which was widely discussed. Their work suggests that the general mood on Twitter predicts the rise or fall of the daily closing price of the Dow Jones Industrial Average with 86.7% accuracy. SNTMNT is strongly influenced by that work, and is another indicator of the growing possibilities in computational social science.
My prediction is that sentiment analysis at either the macro- or micro-level — predicting stock market aggregate moves or the trends for specific stocks — will become a commonplace over the next few years, one of the most obvious applications of big social data.
Now, if someone could only predict the weather…
On June 30th, 2012, in order to keep the time of day close to the mean solar time, a second was inserted into our calendars. So here in Montreal, clocks showed 19:59:59, then 19:59:60, then 20:00:00.
You might guess that a leap second can disturb the highly-dependant-on-time computing world, something like Y2K.
You would be right.
Our analysis systems thus became confused for a while, and new Tweets could not be fetched; fortunately we have recovered most of them, using the current means we have for time travel!
Leap second seldom occur: only happened three times so far in this century. We’re happy to say the underlying technical fix is on its way into our infrastructure.
Thank you for your patience in this matter.
Gossip — talking about people when they are not present — is a staple of human societies, a universal aspect of human interaction. Not surprising, gossip occurs in all social contexts, including online conversations, like email, Twitter, and instant messaging.
“Gossip is a sort of smoke that comes from the dirty tobacco-pipes of those who diffuse it: it proves nothing but the bad taste of the smoker.” – George Eliot
We may consider gossip as negative like George Eliot did, but the anthropological research is fairly consistent in showing that gossip is a necessary to healthy social organizations, whether small or large. Robin Dunbar proposed the idea that human language evolved from the latent desire to gossip (see Gossip, Grooming, And The Evolution Of Language). There is a great deal of research into the exchange of social information that might prove useful, and also the moral implications of the actions of other which can lead to social repercussions for those considered to be bad actors, or untrustworthy.
So, it will come to no surprise that a study of gossip in a large collection of email — the Enron email database of 517,431 messages — shows a consistent set of patterns of gossip. This was reported in Have You Heard?: How Gossip Flows Through Workplace Email, where the researchers — Tanushree Mitra and Eric Gilbert of Georgia Institute of Technology — analyzed the email and found 7.206 messages that they believed were clearly gossip-oriented.
Their findings are very revealing:
“[...] sending email to a small set of people is more frequent and it is more common to see gossip in messages targeted to a smaller audience.
[...] gossip is present in both personal and business email and across all sections of the hierarchy, which demonstrates its all-pervasive nature in organizations.
[...] the hierarchical position of an employee affects his gossip behavior, both in terms of his frequency of gossip and the audience with whom he gossips. Our results indicate that people are most likely to gossip with their peers.
[...] gossip is a social process. Some people are actively involved in generating gossip messages (“gossip source”), while others are silent readers of the messages (“gossip sink”’), and there are some who play both roles.
[...] frequent dyadic email interactions do not show an increase in gossip email. [That is to say that those who are likely to be working more closely may have other opportunities to gossip than email.]“
Other findings are more organizational, like the finding that VPs and Directors at Enron were very likely to move gossip-related emails up to their next immediate supervisor. Also, people at the bottom of the totem pole are most likely to gossip, and to do so among themselves.
This research also suggests that lower tier management is least involved in gossip, although that suggests they may be the topic of greatest gossip, too.
Enron was a disaster as a company, but it may be the case that the patterns of gossip there are perhaps not unusual. The researchers did not delve very deeply into the question of what was being said in these emails, aside from some superficial observations (like the phrase ‘in response to your email’ being common), however, a sentiment analysis suggests that gossip is more strongly correlated with negative emotions (37.44%) than positive ones (13.99%), but most strongly linked to neutral emotions (48.63%). My hunch is the more negative the emotional state of the company, the more negative gossip might arise, but that gossip itself is a kind of background radiation, and is always present.