The field of neural networking is today being explored rapidly. As Stergiou and Siganos (http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html) point out, research had been nearly halted in the field for nearly three decades based on a misconception regarding the ability of a model regarding the mammalian system of vision and its ability to represent different shapes. Until this conceptual problem with the so-called ‘Perceptrons’ was accounted for, the limits of the field were through to be known. Today, on the contrary, the field of neural networking holds the potential to revolutionize the way computer software functions.

Foremost, neural networks are merely modeled after the nervous systems of mammals, and are not meant to mimic their specific function. Instead, the basic principle is borrowed from the biological understanding: in order to generate a desired outcome from a specific set of inputs, a complex intermediary network can use different methods of learning and thereby perform the required behavior. In this way, the network functions much like a nervous system, but these artificial nervous networks are most useful in solving different types of problems than conventional computer algorithms. These networks are instead able to perform primitive types of problem solving for the subjects in which they have been trained as ‘experts.’ Another advantage of this problem solving method is the parallel processes utilized by the network.

The architecture of these neural networks is usually either feedforward or a feedback mechanism. The former, also known as a top-down network, allows for one way flow of data. The more complex feedback networks, however, allow for the network to adapt to its own output, therefore making it approach an equilibrium, at which it remains until disturbed by further input. Neural networks, moreover, are often split into three distinct layers: input, hidden, and output units. Hidden units are essential in the way input information is transformed into output information. The ways in which networks are taught to respond to input are often separated into the classes of associative mapping and regulatory detection.

Neural networks, aside from the computing applications, may be useful in a number of other fields, such as data validation, customer research, and industrial process control. Neural networks can also be very useful in medical settings, where they are used to diagnose patients in a way which is not entirely algorithm dependent. They have even been suggested for use in fields as diverse as marketing and credit evaluation.

http://www.tech-recipes.com/google_tips481.html

The author of the above article discusses the method and success of the marketing scheme Google used for its beta version of GMail, its new email service. Even before the beta version was released in April 2004, news and rumors about a 1-gigabyte mailbox, Google searching emails, and other features were being discussed in online blogs. Once the beta version was released, users could only be invited to use GMail. Therefore, to increase the hype on Internet blogs about the email service, the first invites were sent to Google’s own online bloggers. Then, each new account was given a number of invites to invite their friends to GMail. The number of invites was initially kept low to control growth and keep others eager and excited to get a GMail account. Soon enough, GMail accounts were so coveted that some were even sold for as much as $500 on eBay. This was the essence of GMail’s marketing scheme, to use exclusivity to build hype around their service.

The key to the success in marketing GMail lies in the use of invites to create what Malcolm Gladwell calls a “social epidemic” in The Tipping Point. The idea was to create so much interest and hype around GMail, that this would allow it to market itself. By inviting online bloggers on Google’s blog, GMail took advantage that these bloggers were what Gladwell calls connectors and mavens. These bloggers were considered mavens because they were knowledgeable about GMail and other email users would take their opinions seriously. Furthermore, they were connectors because they were likely, as bloggers, to connect to a large social network of other bloggers and blog readers. These bloggers would then invite friends and fellow bloggers, and soon accounts would trickle down to regular email users. In this sense, the “social epidemic” for the desire to obtain a GMail account began.

Despite some similarity to Gladwell’s “social epidemic” concept, one might think that the chaotic changes he describes in such epidemics were possibly hindered by the limitation of invites. If GMail’s goal was to increase their number of accounts as quickly as possible, shouldn’t they have allowed an unlimited amount of invites? Possibly, but it is possible that such a strategy would cause the increase in accounts to eventually to die down because only those who new about GMail desired one. In effect, GMail would have saturated a large social network. Therefore, by limiting invites, GMail was able to escalate the hype to reach out to other social networks.

Some also argue that GMail also wanted to control the growth of their service to prevent an overload on their servers. Either way, this method was able to accomplish both goals: keep up with the growth and to have hype auto-market GMail.

http://www.scribd.com/doc/5735/Digging-for-Diggers-Analysis-of-a-Social-Media-Website

In response to the other 3 posts about Digg.com, I was able to find a recently posted paper published for a class at Georgia Tech that studies the practices and complaints of the top Digg users. I will assume you know basic knowledge of how Digg works. If not check out the other 3 blog posts. Since the paper is based on interviews with 20 or so of the top Digg users, the paper’s emphasis is on the controversy caused by Digg’s shift of approach when it decided to remove the list of top 100 users in an effort to “balance out the playing field”. Before the removal, users were ranked by a complex algorithm that factored in their popularity (how popular their submissions were), the popularity of their friends, the number of submissions, the number of buries, etc. Since the removal of the list, top users feel as if they’ve been “stripped” of their identity in relation to other digg users, and this has apparently had a negative impact on the social aspect of the site. Diggers are unpaid and one of the main motivators for being a top contributer was the power and recognition. Now, there is less motivation to make friends and interact because one doesn’t know who to befriend in order to gain popularity. On the other hand, Digg has effectively made the site more democratic and in this sense is accommodating for their increasing user population. The content now will be less biased because there will be less “vote trading” among the top users.

One can view this shift of approach as a sustenance of the popular nodes. As a whole Digg is looking to find the best content among the entire community. They have basically eliminated the notion of those who hold monopolies of influence. What has resulted however is a community with two levels of interaction. The top level is the active community - those who have friends on digg, chat via aim, send emails, etc. The other level are the users who go on digg looking for content, burying, digging and who never interact with other users. Digg seems to be reaching out to the second level because that is where the majority of the users are. So although the elimination of the list may seem destructive and not for the greater good, Digg saw many advantages in doing so, which this paper points out clearly. The paper also discusses other interesting issues at hand, which include deviant behavior and the problem of having more and more articles to sift through and sort out in the queue.

…or how you can go from “Network” to “Cadbury Creme Egg” in 12 clicks.

In my search for something “good” to post about I came across some xkcd comics that I couldn’t pass up. Since they don’t really have any intellectual integrity, I thought I would just post them for fun and use one as the basis for my post. Click and enjoy.

Although not the best of the xkcd comics, I found the last one particularly interesting because it is so true. You really can search for something on Wikipedia and before you know it you’ve missed dinner, you’ve learned a whole lot of probably useless information, and you can’t even remember where you started. What is it about that site that allows you get so lost in the web of information? Time to validate the “I’m sorry Professor, I got so lost researching on Wikipedia I couldn’t finish my assignment” excuse and prove a comic to be true all in one fell swoop:

In their study “A Network Analysis of Wikipedia“, Bellomi and Bonato stated:

Wikipedia’s internal references form a single connected graph: there are no separated ”islands” of entries with no inbound or outbound links to the rest of the corpus; given any entry, it is possible to reach any other entry, following a path of undirected reference links.

This can easily be seen in the following images generated on Websites as Graphs thanks to a tip from catrionag (click to enlarge):

The above image shows the network of a search for the term “Networks”. Note the connectedness. So far Bellomi and Bonato’s statement is holding up. There are no “islands”. Everything is connected, even if it’s only by that one lonely node between the two giant clusters. The central nodes, being the pages that are more of references lists, help to connect you from what you were looking for to something completely random and seemingly unrelated.

To test out the theory presented in the xkcd comic, I decided to go on an adventure, click on links within articles, and see where it took me. Within 12 clicks I arrived at “Cadbury Creme Egg”. For comparison, take a look at the following image showing the Websites As Graphs version of the “Networks” page and the “Cadbury Creme Egg” page.

Here we have two (seemingly) distinct things that you would think have very little in common, yet, their page networks are nearly identical. Does this similarity account for the link between the two? Or does the link between the two account for the similarity? Regardless of which, if either, is true, if I challenged you to go from “Networks” to “Cadbury Creme Egg” in those same 12 clicks, odds are it would take you and awfully long time to do it. Go ahead, try it - then tomorrow you can tell your professor you couldn’t finish your homework because you were too busy looking for Cadbury Creme Eggs.

What is one of the most common web sites visited by the students of Cornell University? Facebook.

Facebook is a student social network, that combines to some extent privacy and the ability to socialize in an initially collegiate environment. For example: finding the person that lives in your dorm that sat next you in class and took notes when you fell asleep, but you don’t know on a personal basis. That is just a simple positive example of how a social network can help. However the use of Facebook is limited by its interface and its corporate model.

Facebook will never provide information directly to end user, at least not anymore. All information must occur through some form of Facebook interface that visits their site or a portal through their web page. At the beginning, before it became a business ran by professionals, Facebook used to have the option to export all your friends numbers, emails and contact information to a file that you could import to Outlook and send to your cell phone, to have it all in one place. No more need to stop and look at a computer to find that number. Once it became a corporation however that feature was removed because it limited the use of Facebook as a revenue generating entity; no ads displayed, less profits. This is the major fallacy of online social networks.
The network itself is offered to the end user via a segmented interface that is designed to generate money as well as provide information. But at what price? How much advertising is too much? And how much is it hindering development for the end user?

Myspace and Xanga, two popular blogging and online social networks are even more burdensome to their end users. Myspace provides multiple interfaces that provide data be it via a browser or a cell phone, but at the same time forces much of what is unwanted upon the end user. The page is littered with advertising. Xanga on the other hand uses keyword sensitive advertising that notices what that pages’ blog posts are about.

When presented with large amounts of information the end user should be allowed to filter it in their own individual taste and choice. They are providing specific information to social networks yet receiving back too much or what is unwanted. Filtering the available information from online social networks is prohibitive by their corporate goals. If one end user wanted to scan their Facebook social network and then their Myspace for specific information, they would have to individually visit both rather than be able to develop software that directly accessed the information provided by both networks and scanned for the specific information. The software being the filter is not allowed and limits the usefulness of online social networks.

Facebook as a corporation has realized that it could realize more profits by providing data to developers for such applications. The end user must authorize the software to access their profile from a portal and from that point on it the software can access most of the user information. Though still hindering it does provide the end user more choices in accessing network information. Other social networks should provide similar options.

    Throughout the semester, we have observed many, simplified models of social networks. While these models reduce to people to simple nodes and edges, the main pattern revealed by these models are revealed all around us. I had one such experience when navigating the Ithaca housing market, searching for a place to live next year.

First, let’s take a look at Ithaca’s own Housing Solutions, a website that not only allows one to search through listings, but also tabulates average rents by location and type of apartment. While the information that formed this graph is very complicated, we can easily see how we can apply our Bi-Partite Graph model of real estate to the rents provided. Presumably, people will pay more for more space and less distance from central areas. As one can see, this is reflected in the high prices of larger places to live closer to Cornell and Downtown.

As students, we must also have to form groups to fill up spaces in larger apartments. Because choosing to live is a decision made rarely, it is similar to the experiment we ran when studying power in social networks. In the experimental case, one has to choose to split a dollar before time runs out. In this case, one can only make one or two major decisions a year. Thus students could make power decisions in the same ways the nodes in our experiment did.

Finally, the process of finding a great place to live at a great price sounds like the same kind of information that is spread by the “Strength of Weak Ties.” Because our acquaintances are exposed to a world outside of our own, I have found places to live and new roomates by simply asking others.

The beauty behind the simplicity of the ideas we learn in class lies in their ability to be applied to many situations. While real life situations are often many more times complex, it is exciting to see these broad patterns play out in our daily lives.

Earlier this semester we discussed Vannevar Bush’s article “As We May Think,” which states that the human brain organizes information through associative memory. This idea treats the mind as a web of nodes with connections all over the place. One thing can be connected to a related (or different) topic. For example, the sight of an apple tree could cause someone to think of pies. The manipulation of this web is a goal of advertising. For example, when Tide runs a detergent ad, it hopes that you’ll think of Tide when you see detergent. Who companies target and how they establish connections are marketing challenges. Often, children are targeted through cross promotion.

The goal of cross promotion is to establish links between two brands. This is different from traditional advertising, where the marketer attempts to create a link between a product type and a brand. The benefit of cross promotion can be seen in an example. Let’s say three things cause a person to think of product A and 2 things bring up the thought of B. If a connection can be established between the two brands, then five things can lead to each product. It’s like a merger in the mind.

How children are targeted is described in Eric Schlosser’s Fast Food Nation. He describes Walt Disney as the pioneer in cross promotion because he issued licensing agreements that allowed other companies to use his characters in the 1930s. Walt’s idea continued for the rest of the century, both with his firm and others. For example, McDonalds has teamed up with multiple firms to help sell its Happy Meals. Some of the partnerships have been wildly successful. Schlosser notes that the Teenie Beanie Baby toys increased the sale of Happy Meals by a factor of 10 . Some other cross promotions include Burger King’s partnership with Teletubbies, and Taco Bell’s deal with the NCAA.

Advertisers view children as good targets for cross promotion because “brand loyalty” can be formed early. Schlosser states that “market research has found that children often recognize a brand logo before they can recognize their own name.” If you think bombarding children early to increase the success of a business seems cold, you are not alone. Over the years attempts to limit advertisements have been made. In 1978, the Federal Trade Commission tried to ban child advertising on television. This attempt failed. Others, however, have been more successful. In 2000, the Children’s Online Privacy Protection Act became effective. This legislation limits the amount of information companies can obtain from children under the age of 13.

Link to the Children’s Online Privacy Protection Act:
http://www.coppa.org/#
Link to a review of Fast Food Nation
http://www.mcspotlight.org/media/books/schlosser.html

Schlosser, Eric. Fast Food Nation: The Dark Side of the American Meal. Houghton Mifflin: Boston, 2001. Page 40.
Schlosser, Eric. Fast Food Nation. Page 47.
Schlosser, Eric. Fast Food Nation. Page 48.
Schlosser, Eric. Fast Food Nation. Page 43.
Schlosser, Eric. Fast Food Nation. Page 45-6.

As we have seen in class, graphs of networks can get messy real fast.  There can be hundreds of very small nodes, thousands of edges, different shaped nodes, colored edges, or directional edges all on top of a map of the United States.  It can often take several minutes of staring at a graph to recognize what is going on.  However, these graphs are essential in order to fully understand how a network functions. To avoid this clutter, comprehensive graphs must become even more organized and interactive.  

In a 2004 paper by Viégas and Donath of MIT, the topic of creating better graphs to comprehend ones social network is studied.  The paper can be found here:

<http://alumni.media.mit.edu/~fviegas/papers/viegas-cscw04.pdf>   
 

In their study, Viégas and Donath use two different types of interactive visualizations; Social Network Fragments (SNF) and PostHistory.  The SNF “is a traditional graph visualization that highlights clusters of contacts derived from the TO: and CC: lists in email archives.”  This is the type of social network that we have seen often in class, with people as nodes and friends joined with edges.  The authors describe the PostHistory graph as a metaphor of a calendar which shows how email relationships have evolved over time.  Both graphs contain over a hundred nodes, are very complex, and are shaded in several different colors.  However, the graphs are interactive and the user is able to zoom in on each graph and can actively change the viewing mode.

They found that the user still took some time to get used to the graph, but it also gave the user more insight to their social network than a graph printed on paper.  Additionally, they suggest that these graphs are still too cluttered with a large amount of email addresses shooting from each node.  Viégas and Donath state that “a general neglect of principles of good layout and visual perception on the part of the designers of these systems” is at fault for unorganized graphs.
 

Viégas and Donath also cite the work of Mark Lombardi, who created social network graphs of political scandals and criminal conspiracies.  Several of his works can be viewed here:

< http://www.pierogi2000.com/flatfile/lombardi.html>

Lombardi’s art is quite different from many of the social network graphs we see in class.  Much thought is placed on the position of a node, the curvature of an edge, and the succinct stories that define each relationship.  However, this extreme level of organization does come at a price.  The website above lists one graph as being over 10 feet wide.  This is simply not practical to distribute among a large number of people.
 

Clearly, one of the biggest challenges of creating a good network graph is finding a compromise between conveying the properties of a large network, and showing the detail of the individuals.  Hopefully, future advances in computer programming can help to avoid this dilemma by creating graphs that will not have to compromise between the two. 

In January, there was an interesting post on the Scientific American Observations blog about the “most hated Digg comment.” Commenting on Digg works in much the same way that news stories do. Someone posts the comment, and other users “digg” or “bury” the comment based on whether or not the comment is interesting, relevant, or anything else the user cares about. An interesting property of these comments, as with the news stories, is that once a story becomes popular, more and more people digg it, due to exposure. The stories that get buried usually suffer the opposite fate. They get no exposure, so they just stay at a low number of diggs, and never make it to the main page of Digg.

However, comments that are highly buried do not go unnoticed. Instead of being deleted, they are left on the page, although they become grayed-out, so they do not seem so prominent. But because the comment is still left on the page, it can still be read and evaluated by the users on the page. This happened to an extreme degree in the story highlighted by the SciAm blog. Because a disliked comment happened to be right at the top of the page, thus being viewed by thousands of users, those same thousands often looked at the comment, and decided to go with the majority and bury the comment further.

This says something about the nature of Digg, that has been highlighted by two other previous posts, and goes along with the general “Hubs and Authorities” nature of the Internet, along with the flaws it creates. Just as Google Bombs place irrelevant results at the top of a Google search result page, comments that don’t deserve to be seen by everyone still exist on highly-dugg pages, attracting more attention than should be there.

All in all, the fact that the one comment was buried so slow is due not so much to the fact that Digg users didn’t like it, but rather is due mostly to the underlying network structure of Digg, placing a highly-disliked comment where many are forced into seeing it.

In an attempt to create a new television sensation, and hold on to large numbers of viewers, the FOX television network aired a show entitled Unan1mous last spring, directly after its hit American Idol. This show, which borrowed elements from both Big Brother and Deal or No Deal, featured nine contestants who lived together in an underground bunker. Their sole task throughout the show was to choose which contestant to give 1.5 million dollars to. Every week contestants would vote on who to award the money to and who to eliminate from the chance to win the money. The eliminated players would remain in the bunker and were still counted towards the unanimous vote, but were no longer in the running for the prize.

There were, however, a few catches to the game. The first was that the decision needed to be unanimous, meaning that all contestants had to agree upon one person to award the prize to. The second was that the amount of money continually counted down for every second that the contestants remained in the bunker without coming to a decision. The third was that before each vote an incriminating piece of evidence was publicly announced about a few contestants. Finally, if any contestant chose to leave the bunker, and the game, the money would automatically be cut in half.

The show was filmed over five days. However, the producers had no idea how long it would take the contestants to come to a unanimous decision, and therefore were willing and able to film until a decision was reached or the money ran out. If no one had left the bunker and no decision was reached, the longest the game would have lasted was eight days. Because two players chose to leave, each cutting the pool in half, and a unanimous decision was finally reached, this game did not last the full eight days.

In class we discussed the concepts behind network exchange theory and looked at different examples of power and social exchange experiments. The social network created by this reality TV show is similar to that of “The Ultimatum Game” (if no decision is reached, everyone gets nothing). It is a lot more complicated that the networks we looked at in class but illustrates many of the same principles. Players, in this case, did not have access to full information, as they were allowed to speak to whomever they wished and in many cases were untruthful. One of the contestants chose to lie and told everyone that he had testicular cancer, in hopes that they would feel sorry for him and award him the money. In class we touched briefly on how trust plays a role in social exchange experiments.

We also looked at the Behavioral Principle in class which states that if a node is excluded, it will change what it asks for in order to get back into the game. When the voting results from the entire season of Unan1mous are examined, the majority vote went to a different contestant every round. Instead of trying to convince the few contestants who did not vote for the previous round’s majority winner, contestants picked a new recipient each vote. In hopes to succeed at the game, contestants changed their strategy every round, rather than risk voting for the same contestant they had previously voted for.

Despite the ethical debates surrounding this television show, and the poor cinematography, I find it to be an extremely good example of a social exchange network. I only wish that social scientists had examined the show in further detail because the network of power would be fascinating to graph.