In revisiting the experiments described in my MSc. thesis I discovered that I had made some errors in the computations I had reported. The errors are, to my mind, minor because they do not affect the concept of the type of evaluation I was demonstrating. The concept is as sound as ever. The errors should be corrected nonetheless.
The changes affect Tables 4.1, 4.2 and 4.3 and the minor conclusion in section 5.3.
Idf Finite Paths Finite Correlation 1 -0.7869 0.17% 2 -0.5357 1.45% 3 -0.1271 47.12% 4 -0.1430 78.45% 5 -0.1627 88.72%
Entropy Finite Paths Finite Correlation 1 -0.8099 0.18% 2 -0.5888 0.95% 3 -0.1840 29.97% 4 -0.1483 69.49% 5 -0.1638 81.24%
GfIdf Finite Paths Correlation Finite 1 -0.8050 0.17% 2 -0.6771 0.88% 3 -0.1735 29.49% 4 -0.1544 69.78% 5 -0.1820 84.30%
The conclusion in Section 5.3 `Which Method Is Best' is withdrawn. These corrections do not affect the concept of the type of evaluation.
The IP&M paper I wrote with Bob Webber and Jean Tague-Sutcliffe is mostly about the concept of using a graph to evaluate hypertext link quality.