Recently, two Princeton researchers have applied a variant of the SIR model [irSIR] to explain / forecast the phenomena of "growth" and "decline" of large online social networks like Facebook, Twitter and MySpace, in the same way as you model the phenomena of "infection" and "cure." http://arxiv.org/abs/1401.4208v1
I wonder if this approach is correct.
For completeness, we must add that some FB analysts responded employing the same model, applied to Princeton. With paradoxical results: https://www.facebook.com/notes/mike-develin/debunking-princeton/10151947421191849?utm_content=buffercf163&utm_medium=social&utm_source=twitter.com&utm_campaign=buffer
Dear Giuseppe
Below, the conclusions of the paper:
FIELDS UNDERGRADUATE RESEARCH PROGRAM GROUP REPORT 1
An Epidemiological Approach to Information Propagation in the Digg Online Social Network
Mark Freeman, Harvard University, USA, James McVittie, University of Toronto, Canada,
Iryna Sivak, Taras Shevchenko National University, Ukraine, and Jianhong Wu, York University, Canada
By using a variant of the epidemiological SIR model, we obtained an average predictive accuracy of over 98% for the most popular story, approximately 86% for the 134 highest voted stories and approximately 80% of all stories. These accuracies show that the application of an epidemiological model more accurately predicts the voting trend of Digg network stories than the previous model in [2] over the first 50 hours. In addition to achieving higher accuracies, we showed that it is possible, given restrictive inequalities, to determine the times at which the number of infected users reaches its maximum, the turning point time in the rate of cumulative number of infected and the beginning and ending times of the viral period for any given story. By using these properties from this model, predictions and forecasts can be made which will predict valuable information for those interested in the spread of media on the DOSN
Intuitively I would not be surprised if irSIR is a good predictor. From the conclusion, above, there is evidence that irSIR is a good predictor. Since we are animals and largely behave like animals rather than by using our intellect (a lengthy discussion I don't want to have here) I think a biological model works.
Thanks for sharing the question. May I differ a little while heartily agreeing with all the above about the usefulness of the proposed irSIR model?
Indeed, there are many variants, modifications..etc of the famous SIR Model and the proposed one will certainly not be the last. The original SIR is “born” in the domain of biology while the proposed mutation is led in a very particular social domain that are OSNs and concerns the assumption of a characteristic recovery rate for diseases:
Authors state “intuitively” that [ it is necessary to modify the traditional SIR model to include infectious recovery dynamics] . The proposed model remains as deterministic as the original while its [implication is that there is no way to “immunize” the infected population compartment against infectious recovery in the irSIR model. In the context of the irSIR model, all OSNs are expected to eventually decline.] Yes, not only facebook but all OSNs : I would hope to exclude at least Researchgate:) but seriously was it adequate to compare myspace to facebook : Just a look at http://en.wikipedia.org/wiki/Myspace can bring some light about the VERY different worlds : populations, services, diversities…. and managerial issues. And, yes, there is nothing new in the fact that all products (or services) will end declining some time in future but the managerial role is precisely to delay that time while looking for alternatives including a way to [ “immunize” the infected population compartment against infectious recovery]! OSN are people oriented products (and services) not necessarily “innocent” social phenomena and certainly not biological. Yet, I agree the proposed model IS useful.
I am not aware of the SIR model but do note that use of social networks does resemble addiction a bit. Would SIR also model use of alcohol if suddenly made more available as through supermarkets?
Epidemics in a large population have random and non-random nature.
What is the nature of the spread of Online Social Networks in a large population?
Social networks are virtual communities.
Individuals or organizations join to the network for various reasons:
1. Professional and public (political, cultural) information exchange.
2. Social contacts.
1 is not random. This is obligatory.
2 is the tendency of the information age: instead of personal contacts - virtual contacts. This tendency has a random component.
The spread of epidemics is controlled by the medical service (e., g., mass vaccination).
The spread of Online Social Networks does not controlled.
So the SIR model of the spread of Online Social Networks is only a first approximation.
Joseph, Some mass vaccinations are done with the aim to control the next epidemic.
No, it is not correct in my opinion. The two types of cases are different.
In my opinion .It is not correct because there is a difference between the two types.
In fact, in case of epidemics, the situation is forced on the population as a whole. One falls prey to an epidemic not by design, but by chance. It is another matter that one can perhaps try not to be affected by the epidemic concerned. On the other hand, one joins a social network by design, and not by chance. These two are totally different situations. Using the same mathematical model in these two cases would be illogical.
If a mathematical formula is used in a situation in which it should not be used, the formula is not going to bite the user back! Therefore one is at liberty to use any mathematical model anywhere one likes, but that would not be a scientific analysis anyway.
In epidemiological models, one of the main criteria is to determine the epidemiological threshold: there is an epidemic or not.
In this case [irSIR] model have not such properties to determine and predict jump changes non epidemic -> epidemic.
In fact, the model is a simplified monitoring model for complex nonlinear jump dynamics.
More generalized dynamic models are Markov models.
Hemanta Baruah, you are right! "Everything is correlated with everything".
Ok, the use of the criteria of a SIR is a bit forced.
Nevertheless, Facebook data analyst who disproved the Princeton thesis, forgot to reject not just the model, but the reason for the research.
The reason, in fact, is still untouched:
To consider the use of OSNs as an "epidemic condition", applicable to "a large population". To argue that this is an epidemic - and that, therefore, what we all do every day, several hours a day, should be treated with a modified SIR.
Analysts forgot to defend us, as diseased?
— It must be admitted that, looking at the curve on page 10 of the 2 PhD candidates' paper, that growth÷decrease curve in power law seems exactly the classic one of a long-range epidemiological growth÷decline — say the Oran Plague's one, to be clear…
This way, forcing apart, the thesis designed at Princeton seems pragmatically plausible [any sc. error "acceptable"], and negative for both FB and us.
g
Interesting question though out of my purview. I too come across "decline in Facebook activity", though I have Facebook account I hardly see that. Various opinions are posted. I can say one thing rather than using OSN like Facebook etc.etc it would be better to open some interesting network that can meant only for epidemiological purpose for eg. take GWAS http://gds.nih.gov/ meant for Data sharing typically focus on SNPs.
Googled got basic information from the following
http://wiki.eclipse.org/Introduction_to_Compartment_Models
http://en.wikipedia.org/wiki/Epidemic_model
http://www.unc.edu/~rls/s940/samsidisdyntut.pdf
http://www.bio.utexas.edu/research/meyers/LaurenM/DimitrovINFORMS10.pdf
http://www.tb.ethz.ch/education/model/SIR/sir.pdf
the following article mentioned that models can not completely predict what happens in a given situation
http://arxiv.org/pdf/0910.4443.pdf
interesting link
http://blog.iseesystems.com/modeling-tips/calibrate-models-pest/
the following explained about mixing networks "Networks and epidemic models"
http://rsif.royalsocietypublishing.org/content/2/4/295.full.pdf+html
good chapter about "The Whys, Whats, and Whens of Modelling in Healthcare"
http://www.ams.org/bookstore/pspdf/mbk-74-prev.pdf
The approach in itself may be feasible or not, this is something that I cannot really judge because I do lack the necessary statistical and mathematical skills. If the model yields results that are indeed comparable with the observed patterns in social media, this is a good first-hand indication that it is useful at least. But I think it is necessary to keep in mind that it may be that a superficial resemblance does not capture the essence of what is to be studied, because it is always possible that what is present is a case of parallel evolution. To be precise, a case of different structures producing similar effects out of essentially differing resources to cope with similar external conditions. A good analogy might be the human eye and the eye of the octopus. Therefore I would argue that it is not just a case of something like finding a complex fractal algorithm that produces a pattern matching both epidemics and social networks, as Jon Cloud mentioned. The underlying assumptions of this framework of research seem to me to imply something like meme theory, building a basic analogy between living organisms and, so to say, patterns of information, as the contents of social media are nothing but that (any value discussions aside). That human beings can be hosts to both bacteria and informations should not that easily be taken as an argument to treat both as similar. But perhaps these precautions of mine arise just because I am deeply uncomfortable with the idea of information being alive, and this is another kind of discussion.
A behavioral pattern can spread into a population just like a disease. So, Why not?
DNA, RNA are information at cellular level, so the epidemic pattern is basically the spread of information that triggers a certain behavior. So why not having a more evolved information being transmitted on a completely different level?
On another perspective, social networks compete... so you have an outbreak competing with other outbreaks. Not sure how that fits into the model.
Much more seriously than the 2 "experts" from FB have done responding to the experimental projections of Princeton, I'd report that Suyu Liu, Nicola Perra, Márton Karsai, and Alessandro Vespignani have just published a study applied to the relationship between the network's growth / dynamic / **activities** and epidemiology.
In the explanations of the FB's "experts" [which used the calculations of Princeton paradoxically, sarcastically coming to simulate that our own air to breathe has a limited life (What a find! This is trivially true!)], completely lacked in any element of genetic dynamics, and any correlated predictor.
—g
http://journals.aps.org/prl/abstract/10.1103/PhysRevLett.112.118702
Nice to meet you. I am delighted that a public health model can be tested in a different area other than its disease origins; epidemiology, of course, distinguished public health from any other discipline (See, Drum, Krahn & Bersani on Disability and Public Health, 2009). Of course, Bersani's chapter which relates to over 88,000 governments in the US is promptly dropped at the next gate for epidemiology only.
My background is not "public health" as a discipline (President Obama's International Disability Advisor started with a MPH to honorary doctorate, of course); however, public health departments administer community programs (often in the US, brain injury with medical centers, and home health aides and visitors contracted out to non-MPH agencies or nursing agencies); and have functions which are often community or public administration (never found a public health department that will take a public administration degree yet).
In regard to your "calculations" and "projections", I'd love to review the public health techniques and experiments, which like most community problems, probably have disagreements at the roots versus in the findings. Epidemiology has been very well respected, but the current emphasis of public health has been disputed (from health promotion to addictions and sexual offenders).
Dear Julie,
thank you for your kind reply.
Despite the paper I mentioned does not view me as an author, I'm pretty interested in the similarities between the growth of genetic networks, through which contagions propagate and change in / through different patterns, and the dynamics of genetic networks of another class — not exactly related to the motifs of "public health".
However, my proposal for some reflections is not just a pure comparison between mathematical functions. It is also a sort of concern for the current pervasive identification, by physicists and statisticians, of a model to diagnose the "social media" as a newborn channel for the transmission of diseases.
That is, if now we state that the matter is about diseases, then all of the studies conducted via "social keys" [behaviorism, sociology, psychology — maybe the whole concept of Culture and information] should be transferred toward more appropriate keys of biological manner.
If the intensity and the shape of the information passing through social networks could be framed between the symptoms of some disease, then even an "illiterate" person is a "sick" person; and a very "erudite" person is also very "healthy".
But I still have some doubts.
I look forward to read your review. Best,
—g
Have a look at Vax!, a game about epidemic prevention, using network theory.
—g
http://vax.herokuapp.com/
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning