I’d like to address an interesting topic of online expert networks. For the purpose of this post, the term ‘expert’ is very loosely interpreted to be anyone with significant knowledge of a topic. Such online networks may be blog communities (such as Expert Voices), online forums, IRC groups, mailing lists, and tracing back to the early internet, bulletin board systems (bbs). The formation and stabilization of such networks can be modeled as Professor Easley discussed in class, where each network begins at an unstable equilibrium, with the slight perturbation of ‘more users’ tips the network toward a successful, established and stable equilibrium while the perturbation of ‘less users’ tips the network toward dissolution (and certainly there is a threshold in the definition of ‘more/less’ users).

Online “expert” networks are an amazing resource. There are well established, stable expert networks for most topics one might desire to learn about, and especially discussion related to consumer products (pre and post purchase discussions). For example, if I’m trying to decide what laptop, monitor or external hard drive to buy, I can go to the tomshardware forums and find entire sections devoted to each topic with hundreds of ‘threads’ or topic discussions in each section. The same is true if I’m looking for some great new headphones. I could go to the store try to ask store employee or read the specifications, and I would probably just be horribly confused. I could try to read individual reviews posted by news websites, but I might need to read 30 of them to determine what I’m really looking for. I could also consult an expert network, such as head-fi.org, and find hundreds or thousands of people who discuss the merits of different headphones daily. They also respond to people like me who can post specific requirements and needs. I might even become so involved by making and reading posts on the site that I become somewhat of an ‘expert’ on the topic as well, in which case I can begin contributing to the network as well, thus established forums are self sustaining.

An interesting feature of online forums in particular is that the topics can be renamed and subdivided at will by the administrators of the forum. As such, emerging segregation within the forum can be made explicit by subdividing a topic. For example, computer users who are in a ‘CPU’ discussion may argue that Intel or AMD has the best processor. Initially, the forum might just have a ‘Processors/CPU’ section that includes discussion of Intel and AMD processors. Over time, forum users inevitably join one camp (at least to some extent) and segregate in to ‘AMD’ users and ‘Intel’ users. This leads to the logical creation of an ‘Intel’ subsection and an ‘AMD’ subsection. This is basically an official acknowledgement of cluster formation, although in this case we see a type of ‘hybrid’ cluster, where a user’s ‘affinity’ to a cluster need not be 1 or 0, but some value in between. I may have an affinity of 0.7 to Intel and 0.3 to AMD. This requires the density of the two clusters be defined as the sum of the affinity of each user, and each user may contribute to the density of multiple clusters. In this case, will a user prefer the higher density cluster, the cluster with an affinity closer to his own, or the one to which he has a higher affinity?

This post was prompted by an article written in 1987 about an online expert network of scientists that was used by the courts and government in general for technical advice. Read it here: http://cgi.gjhost.com/~cgi/mt/netweaverarchive/000113.html

 http://www.sbindependent.org/node/1791

As many of us found out, tragedy strikes amongst the Hokies today in Blacksburg, VA. The shooting claimed the lives of 33 people and many are affected by it. Personally, I went to high school in Richmond, VA, and many people I know, friends, schoolmates, mere acquaintances, etc. all are affected.  Many more college and high school students are affected by this event and  than I can ever realize…

Blissfully, I went through the day from class to class, assignment to assignment, not knowing what went on outside of our little Ithaca. I found out about this tragic event through my lab partners this evening and immediately checked Facebook when I got home to make sure everyone I know is okay. As soon as I opened up the homepage, I realized that my minifeed was overwhelmed by the many groups and events that VT Hokies and affected others have put up to not only discuss the event with each other and offer condolences, but also to honor the others.

Social networking web sites like Facebook and Myspace today saved me from insanity and frustration from the vast distance from home. Social networks is an old topic, but still very pertinent today, at least for me.

In the past few classes we’ve been discussing how ideas and preferences travel virally through a population until they hit a cluster. A principle assumption throughout these exercises has been the intelligence and logical behavior of the individual. It’s important that once an individual with better information has spoken out against the trend that the individuals who follow recognize his/her unusual behavior and weight their responses accordingly. However, unlike us, the majority of people are unaware as to how information cascades form, grow, and eventually stop. In fact, you’ve probably seen other’s buying a product just because everyone else has (e.g. the iPod, some fashion trends), after all how can so many people be wrong? Advertisers even have a name for this, it’s called the Bandwagon Effect.

Unfortunately, information cascades are not always as innocuous as product fads. In almost every major conflict or breakdown of society, people have lost their lives for thinking differently to the accepted norm. Breaking an information cascade requires that enough people signal their true beliefs, regardless of the consequences. But that doesn’t happen when you’re likely to get killed or alienated for speaking out; then the cascade becomes an incessant movement, destroying resistance before it can even form.

It’s important for all of us to watch out for negative information cascades and do our best to prevent them.

Stop and think before you help ruin someone’s life by conforming to the majority opinion.

As was previously discussed in class, diffusion in networks is a phenomenon that can be modeled mathematically- by taking a diagram of linked nodes and adding the concept of threshold to it. What happens in some cases (like in the infinite linear chain) is that the cascading effect leads to everyone adopting the idea/trend that is being diffused.

An interesting study at the National Bureau of Economic Research relates information diffusion to stock purchases: Information Diffusion Effects in Individual Investors’ Common Stock Purchases: Covet Thy Neighbors’ Investment Choices. Linking stock purchases with information diffusion, this paper is indeed worth a read.

The factors that drive individual investors to purchase specific stocks is still largely unknown; in knowing more information we can model situations better. The study, which takes information from a real source just like Kossinets and Watts did with emails from Columbia University, through an unnamed “discount broker”, concludes that neighbors living in geographical proximity take to make similar decisions when purchasing stocks.

And just like the people in Columbia University were more likely to know each other if they took classes or had mutual friends near a threshold number, in this case the threshold was physical: 50 miles.

It is very interesting to see how concepts being learned now somehow relate back to what we learned at the start of the semester, and can compliment it nicely.

http://money.cnn.com/2007/03/27/magazines/fortune/fastforward_photobucket.fortune/index.htm?postversion=2007032808

The above article explores the success story behind Photobucket, one of the largest photo-sharing websites currently available. Experts attribute Photobucket’s success to the flexibility it offers when it comes to linking a user’s stored photos. Users can store pictures and short videos on Photobucket and then link them to any other page on the web, including personal blogs and social networking sites such as MySpace and Facebook.

Social networking sites attain success by building communities. They require diffusion in networks to gain membership momentum until a cascade forms, at which point they will blossom. However new competition emerges, just as Facebook came after MySpace to challenge it. Photobucket takes an alternate route: its success is derived entirely from not trying to build a community.

The CEO of Photobucket explains that instead of trying to build a community, they are letting communities such as MySpace and Facebook to build around them. Thus, even if newer social networking communities are created, there is never a need for a user to stop using Photobucket. Again this is due to the flexibility that Photobucket offers. Users invest a lot of time to build their photo albums and would obviously be averse to doing it all over again on a new photo-sharing site. So if they can use Photobucket with all social networking sites, why bother switching? It’s an interesting approach that stands to mitigate the effect of fads and trends.

http://www.nytimes.com/2007/04/14/technology/14DoubleClick.html?_r=1&bl&ex=1176868800&en=a84a2d9cef4524e5&ei=5087%0A&oref=slogin

    Several weeks ago, we talked about the way Google used an algorithm to sell search and contextual based ads to advertisers with a maximum profit. However, there is obviously more to online advertising than just ads that appear based on user inputs, and, as stated in the article, Google has been looking to increase it’s range of advertisements into the realm of display advertisements. By buying DoubleClick, Google has gained control over DoubleClick’s advertisement software, which is especially effective when it comes to display advertising. This is obviously advantageous for Google, but is not the most interesting aspect of their purchase.

            As stated in the article, Google’s purchase of DoubleClick also gave it access to DoubleClick’s “relationships with Web publishers, advertisers and advertising agencies.” This adds a whole new dimension to Google’s ability to sell ads because instead of just relying on mechanical auctions for revenue, Google can now sell ads in something closer to a social network. Assuming that Google plays it’s cards right, it can assume a very powerful position in the network of web publishing and advertising groups and establish business deals that go beyond simple auctions.

Source: Million-Dollar Murray

This article from the New Yorker tells the story of a homeless man in Reno, Nevada. He is constantly being arrested for drunkiness and needs to go to the hospital for various injuries. The story serves as a segway to the nature of homelessness in the United States. The article references a study by Dennis Culhane, who discovered that the time spent in homeless shelters does not follow a normal distribution as previously thought. Instead, it follows the power-law distribution that we studied in class. 80% of people that stay in homeless shelters stay for only a day or two, while 10% tend to return periodically in the winter, and finally the last 10% are the chronically homeless who stay in shelters for extended periods of time. The chronically homeless usually have some kind of disability, and it is this 10% that people normally think of when they think of the homeless.

The power-law distribution of the homeless has important policy implications. Because of the severity of their situation, the chronic homeless consume much more from the government in the form of health care, social-services, and jail time than their counterparts who tend to have shorter terms of homelessness. They tend to visit the emergency room frequently, costing the hospitals thousands of dollars. Occupying one cot at a shelter costs nearly $24,000 each year. Philip Mangano, the executive director of the U.S. Interagency Council on Homelessness, believes in the power-law of the homeless and uses it to reform cities’ policies regarding homelessness. To curb the cost of shelters, he advocates paying extra attention to the poorest of the poor: giving them apartments and checking up on them frequently. This way governments can save money: it is cheaper to pay for an apartment for these people and help them get jobs then to have them on the streets in shelters. This forces the homeless to live within the system and establish themselves. Some people oppose this system; why should the poorest people be rewarded while working mothers are unable to pay rent for a home for their children? The point of this system, however, is mostly to reduce the exorbitant costs that the chronically homeless impose on the system, not necessarily to help those most deserving.

This idea of paying special attention to the few causing the problem can be related to what we talked about in class about the power-law distribution of in-links to web pages. Those web pages with more in-links are probably the more popular sites on the web, and as their popularity grows, more and more of the total bandwidth of the internet will be taken up by these sites. It may be the case that these extremely popular sites will have to be paid special attention to make sure that they do not use up all the bandwidth and drown out the rest of the internet, much like the way the chronically homeless can use up more than their share of social services money.

http://pubs.wri.org/pubs_content_text.cfm?ContentID=575

One of the largest problems facing environmentalists is that of cascade effects surrounding the possible extinction of one species or another. Especially when a “Keystone” predator, one that controls the surrounding population, vanishes, the effect creates a cascade, where the predator’s main pray multiplies and drives out other minor pray, and then in turn consume the natural resources and die out as well.

If you have been reading the news lately, you will have read about Don Imus, a former CBS radio host who was fired last Thursday for calling members of Rutgers women’s basketball team “nappy-headed ho’s.”

What led to his firing? At a superficial glance, perhaps it was completely due to Imus’s bad judgment in using offensive terms to describe a team of young athletes. Certainly, some believe his statement warranted immediate firing. But a recent article in The Wall Street Journal (http://www.freepress.net/news/22398), which describes the sequence of events that cumulated in Imus’s unemployment, shows that information cascades and network effects were also at work.

In the first two days after Imus’s show on Wed., no major media outlet picked up his story, although a liberal media watchdog sent a clip of Imus’s show to hundreds of reporters and the National Association of Black Journalists had denounced his comments in a public statement.

Each reporter who got wind of Imus’s comments had to make a decision about whether or not to cover the story based on his/her signal of whether or not this was a newsworthy story. In the beginning, reporters were getting “low” signals, but on Friday, after Imus made a public apology on his radio show, the signal switched to “high.” A cascade started, and reporters began calling CBS and MSNBC in large numbers.

While the media coverage was taking off, CBS and MSNBC still were trying to decide on a course of action. CBS had initially planned to suspend Imus for only two weeks. An executive overseeing some of the radio stations airing Imus’s program said that ‘“nobody had a firm hand on it,”’ and local stations said “the situation was drifting,” according to the WSJ. At this point, I believe Imus’s situation could have worsened or improved – either was a possibility.

Then, opinion-leaders began to criticize Imus. The Reverend Al Sharpton and Jesse Jackson organized rallies over the weekend, and Procter & Gamble, one of CBS’s most conservative advertisers, began to talk of pulling their advertising. I would argue that after these leaders became involved, Imus’s fate was sealed. Many other advertisers decided to pull their advertising after P&G announced its decision, even though some had not planned to. P&G’s action had a network effect – it decreased the payoff for the remaining advertisers in CBS’s network. From the point of view of an individual advertiser, the risk of being the only advertiser “supporting” Imus’s remarks increased, and each company wanted to pull out so it wouldn’t be the only advertiser left. In addition, P&G’s action could have helped to start an information cascade. Each advertiser had its own information on how much sticking with CBS would damage public relations. Perhaps most advertisers thought that it would only damage them a little. However, seeing the actions of P&G and one or two other companies, the remaining advertisers that had low signals would infer that the ones who pulled out had a “high” damage signal, and so they rationally followed these companies.

The other interesting thing about Imus’s story is that he went onto Sharpton’s show in an attempt to apologize, but he only succeeded in fueling the flames. To extrapolate a bit on the discussion of network clusters we had in class, we can see that Al Sharpton is a focal point in a cluster sharply opposed to Imus, and he might have had few friends outside of this cluster. Within this network perspective, Imus had little chance of getting Sharpton to “adopt” an alternative viewpoint. Perhaps a more effective strategy for Imus would have been to find supporters farther away from the cluster’s center (who might have a larger percentage of friends outside the cluster), or supporters outside the cluster but still with many friends in the cluster. Imus could also have tried to enlist the support of influential people, such as a celebrity, to fight against the network effects of Sharpton and P&G.

Without any judgment about whether Imus deserved to get fired, what do you think Imus could have done to tip the situation in his favor and save his job?

These two papers are pretty similar and are written by the same author, W. Brian Arthur: Positive Feedbacks in the Economy,

Competing Technologies, Increasing Returns, and Lock-In Historical Events. Both papers are from the late ’80s, and show that the idea of Network Effects dates back, although in these paperes it is discussed in terms of positive feedback. The paper mentions the example that Professor Easley gave of the classic Betamax versus VHS adoption, where it is argued that it is not superior technology that won out, and that the outcome was influenced by luck. In a sense, this also is symmetry breaking, as “Both systems were introduced about the same time and thus started with roughly equal market shares, but those shares fluctuated early on due to external circumstance, “luck” and other actions by companies manouvering for position.” The paper talks about two positive feedback loops which we have discussed as network effects,”the more people adopt a particular technology, the more it improves, and the more incentive there is for further adoption,” and, “increasing adoption makes it advantageous for newcomers to a field–who must exchange information or products with those already working in the field–to fall in with the standard.”

These papers also combine some of the earlier topics, such as graph theory and the importance of building ties, “…hi-tech products, unlike bulk goods, require specialized marketing and good relationships with customers. Increasing market share requires building a network of such ties; the more extensive this network the easier further increases become.”

Also interesting in the papers is seeing the nascence of this analysis and the quasi-prediction that it would become more common as technology develops, “If increasing-returns mechanisms are important, why have they been largely ignored until recently? Some would say that complicated products–high technology–for which increasing returns are so prevalent, are themselves a recent phenomen. This goes back to what was said in class that the very low marginal cost involved is also important to creating network effects, where distributing more units of already programmed software can be simplified to having no marginal cost.