Traffic-based feedback on the web

Abstract

Usage data at a high-traffic web site can expose information about external events and surges in popularity that may not be accessible solely from analyses of content and link structure. We consider sites that are organized around a set of items available for purchase or download, consider, for example, an e-commerce site or collection of online research papers, and we study a simple indicator of collective user interest in an item, the batting average, defined as the fraction of visits to an item's description that result in an acquisition of that item. We develop a stochastic model for identifying points in time at which an item's batting average experiences significant change. 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.

Temporal Dynamics of On-Line Information Streams

“…In joint work with Jon Aizen, Dan Huttenlocher, and Tony Novak [3], the author studied the dynamics of download behavior at one such site, the Internet Archive, which maintains a publically accessible media collection consisting of old films, live concerts, free on-line books, and other items available for download. The basic definition in [3] is the "batting average" (henceforth, the BA) of an on-line item: the number of acquisitions divided by the number of visits to the description page.…”

“…The basic definition in [3] is the "batting average" (henceforth, the BA) of an on-line item: the number of acquisitions divided by the number of visits to the description page. Thus a high BA is a reflection of item quality, indicating that a large fraction of users who viewed the item's description chose to download it.…”

“…Such an instantaneous BA must be synthesized from the 0-1-valued sequence of user download decisions: each entry in the usage log simply reflects whether a particular chose to download the item or not. In [3], two different methods are considered for determining an instantaneous BA from this sequence. First, one can apply Gaussian smoothing, computing an average of each point in the download sequence with its neighbors, weighted by coefficients that decay according to a Gaussian function.…”

Data association for topic intensity tracking

“…Furthermore, it is not easy to select the appropriate length for the time window, since, depending on the topic intensity, the same time window will contain very different numbers of messages. Also, from the perspective of identifying bursts in the data, a set of discrete levels of intensity is preferable (Aizen et al, 2004). To overcome these problems, Kleinberg (2003) proposed a weighted automaton model (WAM), an infinite-state automaton, where each state corresponds to a particular discrete level of intensity.…”

A framework for analysis of dynamic social networks

“…[14]. The last few years have seen a development of systematic algorithmic approaches to dynamic network analysis, mostly in the context of information networks [1,18,20,19,23,24]. Yet, most of the methods focus on the frequency, rather than concurrency and order of interactions.…”