View source for Gruhl et al www 2004

From Cohen Courses

← Gruhl et al www 2004

You do not have permission to edit this page, for the following reason:

---

You can view and copy the source of this page.

== Citation == '''Daniel Gruhl , R. Guha , David Liben-Nowell , Andrew Tomkins, Information diffusion through blogspace, Proceedings of the 13th international conference on World Wide Web, May 17-20, 2004''' == Online Version == http://dl.acm.org/citation.cfm?id=988739 == Summary == This [[Category::paper]] characterizes information diffusion through blogspace along two dimensions: topics and individuals. The model assumes that topic distributions are mainly comprised of ''chatter'' and ''spikes''. The paper explores snapshot topic models which focus on short term behavior (weeks or months). The features for topic extraction were basically proper nouns that were considered interesting using [[UsesMethod::tf-cidf]]* with a threshold of tf > 10 and tfcidf > 10. With the extracted words, they manually identified 340 classical topics. They identify 3 types of topic patterns : [[File:Spiky.png]] * Just Spike: Topics which became suddenly active and went back to being inactive again. * Spiky Chatter: Topics that react quickly and strongly to external events and have many spikes. * Mostly Chatter: Topics which are continuously discussed at relatively moderate levels. By characterizing these patterns qualitatively, they were able quantify each topic with two parameters corresponding to the chatter level and the spike pattern. Their results show that their intuition of spikes is well captured by the model. The next part focuses on modeling how a given topic spreads over various individuals. They use a few predicates on topics using which they use to represent life-cycle of a topic. The table below contains these predicates. [[File:Lifecycle.png]] The model is derived from [[UsesMethod::Independent cascade model]] of Goldenberg et al. The interesting results from this analysis are : * Most users leave the topic with less energy than it arrived, transmitting to less than one additional person. * Some users provide a boost to every topic they post about. * Also they report that there are critical linkages, some individuals tend to be more strongly associated with topics than others. '''chatter''' Ongoing discussion whose subtopic flow is largely determined by the authors. '''spikes''' Short-term, high intensity discussion of real-world events that are relevant to the topic. *tf- Term Frequency, cidf - Cumulative Inverse Document Frequency. == Dataset == The corpus was formed by collecting daily crawls of 11,804 blog feeds. 2K - 10K blogs posting per day. Also for the corresponding period, data from 14 RSS channels of rss.new.yahoo.com were collected. This was to keep track of what was going on in the media. They used [http://en.wikipedia.org/wiki/IBM_WebFountain WebFountain] to store the data as parent/child entities. == Study Plan == For Topic Modeling * Sequential Pattern Mining [http://rakesh.agrawal-family.com/papers/edbt96seq.pdf] * http://dl.acm.org/citation.cfm?id=873830 For Individual Modeling * http://www.complexmarkets.com/files/TalkofNetworks.pdf == Related Work == * Norman T. J. Bailey. The Mathematical Theory of Infectious Diseases and its Applications. Griffin, London, 2nd edition, 1975. * Cristopher Moore and M. E. J. Newman. Epidemics and percolation in small-world networks. Physical Review E, 61:5678–5682, 2000. cond-mat/9911492. * M. E. J. Newman. The spread of epidemic disease on networks. Physical Review E, 66(016128), 2002. cond-mat/0205009.

Return to Gruhl et al www 2004.