Friday, November 21, 2008
The Napoleon Dynamite Problem
I was not planning on blogging at the end of week ten but this article in the New York Times caught my eye. It discusses the Netflix challenge to create an algorithm to improve their recommendation algorithm. The part that I found particularly interesting is that certain movies are really hard to classify. Bertoni, the computer programmer in the article, says that his algorithm is really accurate for the fast majority of movies but there are a few movies that are really hard to predict. Napoleon Dynamite for example. It is one of those movies that people either love or hate and it is hard to say why. He says that other polarizing movies such as Lost in Translation, Fahrenheit 9/11, and Kill Bill are also hard to predict. The difficulty of predicting specific movies and relative ease of predicting others adds an interesting dynamic to the network of movie preferences.
Tuesday, November 11, 2008
Modeling the Flu With Google Search Queries
There is an interesting article in today's NYTimes that outlines how Google is using search queries to model flu outbreaks. Google Flu Trends watches for key search terms that could indicate someone has the flu: thermometer, flu symptoms, muscle aches, congestion, etc. By knowing when and were the searches originate they can model the spread of the flu.
They have tested their models against Center for Disease Control (CDC) data--they claim they can see the start of a flu strain 7-10 days before the CDC. Google has also tested their historical data against the CDC's and found very high correlations. There is a study in the works that will be published in Nature.
From the article:
“This seems like a really clever way of using data that is created unintentionally by the users of Google to see patterns in the world that would otherwise be invisible,” said Thomas Malone, a professor at the M.I.T Sloan School of Management. “I think we are just scratching the surface of what’s possible with collective intelligence.”
Oh, and our friend Hal Varian is quoted in the piece.
They have tested their models against Center for Disease Control (CDC) data--they claim they can see the start of a flu strain 7-10 days before the CDC. Google has also tested their historical data against the CDC's and found very high correlations. There is a study in the works that will be published in Nature.
From the article:
“This seems like a really clever way of using data that is created unintentionally by the users of Google to see patterns in the world that would otherwise be invisible,” said Thomas Malone, a professor at the M.I.T Sloan School of Management. “I think we are just scratching the surface of what’s possible with collective intelligence.”
Oh, and our friend Hal Varian is quoted in the piece.
Friday, November 7, 2008
Obesity: A contagious disease?
It seems obvious that diseases such as colds and STDs are spread through social networks, but it now seems that obesity may also be. The Framingham Heart Study, a longitudinal study looking at health, specifically weight, among people and their friends, shows that obesity tends to cluster in social networks.
It does not seem surprising that obese people may tend to be friends with other obese people, but the Framingham study shows that if someone becomes obese in a given time interval, their friends have a greatly increased chance of also becoming obese. Possible reasons for this could be that friends may simultaneously adopt similar lifestyle choices such as diet, exercise or smoking, which could effect their weight. There could also be changing attitudes towards weight that could spread through social networks and make one more or less inclined to gain/lose weight because of social pressures. Similar effects were seen among siblings, where if one gained weight, another sibling is also likely to gain weight.
The effects both among siblings and friends are strongest among same sex relationships. They also depend primarily on social distance, ie. closer friends have stronger impacts, and the effects appear to be independent of geographic distance.
Another article on the spread of obesity through social networks, looks at how obesity, and other diseases, should be looked at as a hierarchy of networks (see figure below). On the bottom, there are all of the molecular networks such as gene regulation, protein interaction and metabolic networks which can be studied to look at the molecular basis of disease. Then the paper discusses a disease network where diseases such as obesity are also linked to other diseases such as heart disease and diabetes. Diseases could be connected because they frequently co-occur, one leads to another, they involve similar molecular components or other types of connections. Lastly, there is the social network. Diseases such as the flu or HIV could be passed directly through a social network or it could be a transfer of attitudes or lifestyle.
The relation between the networks adds an interesting angle to the subject. For example, genetics. One's genetics most obviously effect the molecular aspects of disease, but the social network has a big impact on genetics because within an obesity cluster, genetic predispositions for obesity are likely to be passed on to children from both parents. The combined factors of genetics, lifestyle and social pressures could make these children highly prone to obesity.
The interacting levels of networks may also lead to new strategies for treating obesity. The Framingham study emphasizes that their findings suggest that the social network provides an important resource in treating obesity as well and that perhaps social support groups could be a helpful strategy. It seems so human ecological to say it, but issues like obesity and other public health problems, need to be addressed on many levels, and the hierarchy of networks provides a model for how those levels interact.
It does not seem surprising that obese people may tend to be friends with other obese people, but the Framingham study shows that if someone becomes obese in a given time interval, their friends have a greatly increased chance of also becoming obese. Possible reasons for this could be that friends may simultaneously adopt similar lifestyle choices such as diet, exercise or smoking, which could effect their weight. There could also be changing attitudes towards weight that could spread through social networks and make one more or less inclined to gain/lose weight because of social pressures. Similar effects were seen among siblings, where if one gained weight, another sibling is also likely to gain weight.
The effects both among siblings and friends are strongest among same sex relationships. They also depend primarily on social distance, ie. closer friends have stronger impacts, and the effects appear to be independent of geographic distance.
Another article on the spread of obesity through social networks, looks at how obesity, and other diseases, should be looked at as a hierarchy of networks (see figure below). On the bottom, there are all of the molecular networks such as gene regulation, protein interaction and metabolic networks which can be studied to look at the molecular basis of disease. Then the paper discusses a disease network where diseases such as obesity are also linked to other diseases such as heart disease and diabetes. Diseases could be connected because they frequently co-occur, one leads to another, they involve similar molecular components or other types of connections. Lastly, there is the social network. Diseases such as the flu or HIV could be passed directly through a social network or it could be a transfer of attitudes or lifestyle.
The relation between the networks adds an interesting angle to the subject. For example, genetics. One's genetics most obviously effect the molecular aspects of disease, but the social network has a big impact on genetics because within an obesity cluster, genetic predispositions for obesity are likely to be passed on to children from both parents. The combined factors of genetics, lifestyle and social pressures could make these children highly prone to obesity.The interacting levels of networks may also lead to new strategies for treating obesity. The Framingham study emphasizes that their findings suggest that the social network provides an important resource in treating obesity as well and that perhaps social support groups could be a helpful strategy. It seems so human ecological to say it, but issues like obesity and other public health problems, need to be addressed on many levels, and the hierarchy of networks provides a model for how those levels interact.
Sunday, November 2, 2008
Readers of Taste
"The main problem, if that's the word, is that we live in the physical world," writes Chris Anderson. The problem, if that's the word, is "suffering the tyranny of lowest-common-denominator fare, subjected to brain-dead summer blockbusters and manufactured pop." And to free ourselves of this tyranny is to see expression of "our true taste, unfiltered by the economics of scarcity." The Internet's virtuality, then, offers the ideal, trapped in the physical world, unmediated expression. Anderson gets palpable, winking glee at the coming of this utopia from the possibility that we may all be individuated by consumer choices. He would find companions among the economists who, looking only at the numbers, argue that oil is a commodity only limitable by technologic development, which they see as finding its limitlessness through absolution from physical constraints. The physical medium is ignored in favor of its purpose: in Anderson's case, to buy digital products. There is an infrastructure at work here, whose costs are becoming more and more negligable, but is dependent on not just fallable sytems, but on human labour. The college student downloading iTunes tracks may do so because of a physical network that must be maintained and powered. "The Long Tail" is only long in a particular dimension; Anderson neatly only addresses anyone who cares to and can read his article by referring to us as "we."
Thursday, October 23, 2008
"Do Schools Kill Creativity?" Well, find that out!
I know this has nothing to do with the class but you should watch it anyways. Nafisa and I were surfing on TED.com and we found Sir Ken Robinson's talk on "Do Schools Kill Creativity?" Why should you watch this? He is funny; mentions human ecology all the way at the end; points some fascinating thoughts about "you" and the creativity.
It is weeks six and you need a little break. And of course, you should use your imagination to create a network niche with the topic. Trust me you wont regret! Here is the video. comments are welcomed.
It is weeks six and you need a little break. And of course, you should use your imagination to create a network niche with the topic. Trust me you wont regret! Here is the video. comments are welcomed.
Binomial Coefficients and the Election
I've been following 538.com: Electoral Projections done Right very closely the last few weeks. The blog authors collect polling data from a wide range of sources, weigh it according to their sense of the poll's reliability, and produce aggregate projections. I'm not an expert, but from what I can tell the blog's authors are very good; I've learned a lot from reading the blog and I think their way of handling data makes a lot of sense. If you're interested in the art and mathematics of political polling, or if you just want to follow the build-up to the election, I highly recommend it.
Anyway, just moments ago what's wrong with this picture appeared in my feed reader from 538.com. In it, the author skewers some fishy poll results that were published earlier today. To do so, he makes reference to the binomial distribution. Proof that I wasn't lying when I said in class that binomial distributions are super useful.
Anyway, just moments ago what's wrong with this picture appeared in my feed reader from 538.com. In it, the author skewers some fishy poll results that were published earlier today. To do so, he makes reference to the binomial distribution. Proof that I wasn't lying when I said in class that binomial distributions are super useful.
Network analysis of the IPCC
Last class we briefly discussed how networks could be applied to a vast array of topics. This term in my Global Environmental Politics I’ve developed a strong interest in the politics of climate change. A few weeks ago I stumble across a paper on the “Network Analysis of the Intergovernmental Panel on Climate Change (IPCC),” in which Travis Frank, Robert Nicol, and Jaemin Song from MIT examine the team structure, network architecture, and other major influences of the IPCC’s Third Assessment Report.
The IPCC–who in 2007 shared the Nobel Peace Prize with Al Gore– is the scientific body in charge of assessing on a comprehensive, objective, open and transparent basis the latest scientific, technical and socio-economic literature produced worldwide relevant to the understanding of the risk of human-induced climate change, its observed and projected impacts and options for adaptation and mitigation.
The hundreds of climate change scientists who were members of the IPCC at that time, their nationality, fields of expertise, team collaboration among other things make this study quite interesting in understanding the politics of climate change within the scientific community.
The calculation of the clustering coefficient shows that the authors in the IPCC report are less connected than Newman’s least connected research field. They also found that the “longest shortest” path in the network was 19 edges. “Centrality betweenness” is used to rank the top 20 authors and shows that developing countries are not represented as well as the developed countries.
I feel that this paper is a clear example of how network theory can be applied to a specific area to prove arguments related to collaboration in climate change.
The IPCC–who in 2007 shared the Nobel Peace Prize with Al Gore– is the scientific body in charge of assessing on a comprehensive, objective, open and transparent basis the latest scientific, technical and socio-economic literature produced worldwide relevant to the understanding of the risk of human-induced climate change, its observed and projected impacts and options for adaptation and mitigation.
The hundreds of climate change scientists who were members of the IPCC at that time, their nationality, fields of expertise, team collaboration among other things make this study quite interesting in understanding the politics of climate change within the scientific community.
The calculation of the clustering coefficient shows that the authors in the IPCC report are less connected than Newman’s least connected research field. They also found that the “longest shortest” path in the network was 19 edges. “Centrality betweenness” is used to rank the top 20 authors and shows that developing countries are not represented as well as the developed countries.
I feel that this paper is a clear example of how network theory can be applied to a specific area to prove arguments related to collaboration in climate change.
Subscribe to:
Posts (Atom)