Mapping Knowledge Domains (2004) / Chapter Skim
Currently Skimming:
Traffic-based feedback on the web
Traffic-based feedback on the web
Pages 72-78
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 72... ...
In experiments with usage data from the Internet Archive, we find that such changes often occur in an abrupt, discrete fashion, and that these changes can be closely aligned with events such as the highlighting of an item on the site or the appearance of a link from an active external referrer. In this way, analyzing the dynamics of item popularity at an active web site can help characterize the impact of a range of events taking place both on and off the site.
|
From page 73... ...
Because its media collections became public, the Internet Archive site has featured continuously updated lists of the items with the highest acquisition count, displayed separately for movies, audio, and texts. Beginning in November 2002, on the same pages, lists of the items with the highest batting averages (corrected for small sample sizes)
|
From page 74... ...
The rationale for the two-slope model underlying the state transition cost is that we want to be able to detect sudden large changes in bias that result from discrete external events such as adding a new link, while at the same time not having the model drift from one bias to another and back again. Thus the higher slope for small changes encourages the system to stay in a given state rather than bouncing around, but at the same time large changes in bias do not incur much more of a penalty than moderate changes.
|
From page 75... ...
In summary, the example in this section suggests that an approach based on a HMM with a large number of underlying states can accurately localize points of discrete change and can capture changes in interest in an item over time scales that range from hours to months in duration. Aligning Changes in Interest with External Events The crux of the example in the previous section was that significant changes in the batting average for an active item are often correlated with "real-world events," both on and off the site, in which this item is featured.
|
From page 76... ...
The effect of condition i is to require the transitions to localize events closely in time; the effect of condition ii is to prevent a single observed event from "explaining" multiple major transitions. Because the ideal is for major transitions to lie near observed events, we will say that such an alignment accounts for the transitions rat, r2, .
|
From page 77... ...
Further Directions A site as active as the Internet Archive has events of many different kinds impinging on it simultaneously: users view and download items, write reviews, and post messages to discussion boards; active external sites discuss the archive and drive traffic to it; world events generate interest in particular items at the archive. Our probabilistic model for identifying changes in the batting average allows us to analyze one of these streams of actions, the sequence of download decisions, in a principled fashion.
|
From page 78... ...
(2000) in Proceedings of the IEEE Computer Vision and Pattern Recognition Conference, eds.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.