|
|
The Flow of Information in Networks
Abstract: The information we experience
online comes to us continuously over time, assembled from many small pieces,
and conveyed through our social networks. This merging of information,
network structure, and flow over time requires new ways of reasoning about
the large-scale behavior of information networks. I’ll discuss approaches
for tracking information as it travels and mutates in online networks,
showing how to capture temporal patterns in the news over a daily
time-scale – in particular, the succession of story lines that evolve,
compete for attention, and collectively produce an effect that commentators
refer to as the news cycle. I’ll also discuss how to quantify the influence
of individual media sites on the popularity of stories and an algorithm for
efficiently finding influential media sites.
This talk includes joint work with
Lars Backstrom, Jon Kleinberg and JaewonYang.
|
|
Tentative topic: “Inferring relevant networks from
social communication data”
|
|
The Breadth of Contagion of the
Oprah Effect:
The Diffusion of Demand Shocks in a Recommendation
Network
Abstract: A number of offline media
events have a significant impact on online consumer choices, demand and
search. Moreover, owing to the hyperlinked nature of the web (and
specifically of e-commerce web sites), the span of impact of these offline
events is not restricted to the topics or products that were their original
focus. Rather, the “shocks” created by these events spread through networks
of interconnected products, pages or people. In this study, we investigate
the online contagion of exogenous demand shocks created by offline media
events. Our context for offline media events are book reviews featured on
the Oprah Winfrey television show and in the Sunday New York Times. We studied
the impact and diffusion of these exogenous events on the demand for a
“network” of related books that were not explicitly mentioned in a review
but were located “close” to a reviewed book in an online network. This
network is a visible co-purchase network created based on historical demand
patterns at a leading online retailer during the years 2006-2007.
Using a
difference-in-differences matched sample approach we identified the extent
of the variations caused by the visibility of the online network (i.e., by
consumers clicking on visible hyperlinks) and distinguished this effect
from variation caused by hidden product complementarity. We found a
strikingly high level of diffusion of exogenous shocks through such
networks. Neighboring books (even four clicks away) experienced a dramatic
increase in their demand levels, even though they were not actually
featured in the review; this effect is indicative of the depth of contagion
in online recommendation networks following exogenous shocks.
We found that product
characteristics, assortative mixing and local network structure play an
important role in explaining which books will be affected by the shock, as
well as, the relative persistency of the shocks. Most interestingly, we found
that clustered networks “trap” a higher fraction of the contagion closer to
the reviewed book. Our research provides an important documentation of the
magnitude and persistence of diffusion of demand shocks across product
networks, and evidence of the important role and influence of product
networks in electronic commerce in the presence of exogenous shocks.
Joint work
with Eyal Carmi (Tel Aviv University) and Arun Sundararajan (New York
University)
|
|
The group-first approach to social churn prediction
|
|