PDF Publication Title:
Text from PDF Page: 167
7.1.3 Why use such a strict tolerance in your computation? In all the PageRank computations throughout the thesis, we computed PageRank vectors with a strict tolerance, typically tighter than 10−10. These vectors are needlessly accurate. Many applications use PageRank vectors with loose tolerances around 10−4 [Kamvar et al., 2003]. We felt that computing PageRank accurately was necessary to distinguish between the effects of our new sensitivity measures and the effects due to inaccurate computations. Many of the PageRank values are small and a few are quite large. Distinguishing differences among the small values implies we should use a strict tolerance. Also, the real answer to the tolerance question is: because we can. The graphs studied in this thesis are small compared with industrial web graphs. In this sense, the difference in computation time for extra accuracy is mean- ingless. There are no application requirements we have to meet, so why not get extra accuracy? 7.1.4 What about ties in the PageRank vector? At various points in this thesis, we illustrate a PageRank vector with an ordered list. For the nodes with highest PageRank, showing an ordered list is okay because the top few PageRank values are clearly separated from each other. The remainder of the PageRank vector, however, is often riddled with tied values. These ties are identical floating-point numbers and not just values within the machine precision tolerance. Exact floating-point ties occur when two pages have identical in-links, the value of v is the same on both pages, and the PageRank solver is invariant to permutations.2 While we do not attempt to quantify the total number of ties—they do seem to be common. This affects the results here in two ways. First, the Kendall-τ computation requires the order of the nodes. We used a version of τ that incorporates tied values, however. Second, the intersection similarity measure also uses a ranked order. This computation may change in the presence of ties. We only expect a slight change as the measure itself is considerably less sensitive to tied values. This follows because of the smoothing effect in the running average nature of the metric. In short, ties are a problem with some PageRank computations, but we do not expect them to alter the results of this thesis in any meaningful way. 2 It is worth noting that Gauss-Seidel algorithms are not invariant to permuta- tions. This may suggest that they are less reliable for ranking purposes. 7.1 ⋅ discussion 147PDF Image | ALGORITHMS FOR PAGERANK SENSITIVITY DISSERTATION
PDF Search Title:
ALGORITHMS FOR PAGERANK SENSITIVITY DISSERTATIONOriginal File Name Searched:
gleich.pdfDIY PDF Search: Google It | Yahoo | Bing
Cruise Ship Reviews | Luxury Resort | Jet | Yacht | and Travel Tech More Info
Cruising Review Topics and Articles More Info
Software based on Filemaker for the travel industry More Info
The Burgenstock Resort: Reviews on CruisingReview website... More Info
Resort Reviews: World Class resorts... More Info
The Riffelalp Resort: Reviews on CruisingReview website... More Info
CONTACT TEL: 608-238-6001 Email: greg@cruisingreview.com (Standard Web Page)