Comments on: Predating Link Models and PageRank http://irthoughts.wordpress.com/2008/02/14/predating-link-models-and-pagerank/ News, Papers, and Theses on Information Retrieval, Data Mining, and Search Engine Technologies. Fri, 25 Jul 2008 15:29:27 +0000 http://wordpress.org/?v=MU By: E. Garcia http://irthoughts.wordpress.com/2008/02/14/predating-link-models-and-pagerank/#comment-539 E. Garcia Fri, 15 Feb 2008 12:55:20 +0000 http://irthoughts.wordpress.com/?p=168#comment-539 Thanks, snunes, for stopping by. <a href="http://research.microsoft.com/~najork/sigir2007.pdf?0sr=p" rel="nofollow">Najork's paper at 2007 SIGIR</a> suggests that link-spammers targeting PageRank may play a role in their finding. <blockquote> The fact that features based on outgoing links underperform those based on incoming links matches our expectations; if anything, it is mildly surprising that outgoing links provide a useful signal for ranking at all. On the other hand, the fact that in-degree features outperform PageRank under all measures is quite surprising. A possible explanation is that link-spammers have been targeting the published PageRank algorithm for many years, and that this has led to anomalies in the web graph that affect PageRank, but not other link-based features that explore only a distance-1 neighborhood of the result set. Likewise, it is surprising that simple query-independent features such as in-degree, which might estimate global quality but cannot capture relevance to a query, would outperform query-dependent features such as HITS authority scores.</blockquote> I look forward to AIRWeb 2008 to peer review manuscripts along those lines or at least read what others conducting link-based spam research are going to submit. On the other hand, Najork et. al. used breadth-first crawling to collect links and phantom nodes to avoid the sink problem. It would be interesting to see how their results compare if these experimental conditions are changed. Thanks, snunes, for stopping by.

Najork’s paper at 2007 SIGIR suggests that link-spammers targeting PageRank may play a role in their finding.

The fact that features based on outgoing links underperform those based on incoming links matches our expectations; if anything, it is mildly surprising that outgoing links provide a useful signal for ranking at all. On the other hand, the fact that in-degree features outperform PageRank under all measures is quite surprising. A possible explanation is that link-spammers have been targeting the published PageRank algorithm for many years, and that this has led to anomalies in the web graph that affect PageRank, but not other link-based features that explore only a distance-1 neighborhood of the result set. Likewise, it is surprising that simple query-independent features such as in-degree, which might estimate global quality but cannot capture relevance to a query, would outperform query-dependent features such as HITS authority scores.

I look forward to AIRWeb 2008 to peer review manuscripts along those lines or at least read what others conducting link-based spam research are going to submit.

On the other hand, Najork et. al. used breadth-first crawling to collect links and phantom nodes to avoid the sink problem.

It would be interesting to see how their results compare if these experimental conditions are changed.

]]> By: snunes http://irthoughts.wordpress.com/2008/02/14/predating-link-models-and-pagerank/#comment-538 snunes Fri, 15 Feb 2008 11:42:56 +0000 http://irthoughts.wordpress.com/?p=168#comment-538 What do you think of the findings made by Najork et al. (2007) that conclude that HITS is not very different from a simple inlink count? Web-based features are theoretically very robust but it seems that in practice the results are not as good as expected. Content-based features are much more interesting. "We were quite surprised to find that HITS, a query-dependent feature, is about as effective as web page in-degree, the most simpleminded query-independent link-based feature." in HITS on the Web: How does it Compare? http://research.microsoft.com/research/pubs/view.aspx?0rc=p&type=Publication&id=1734 Great blog! Thanks for your posts. What do you think of the findings made by Najork et al. (2007) that conclude that HITS is not very different from a simple inlink count? Web-based features are theoretically very robust but it seems that in practice the results are not as good as expected. Content-based features are much more interesting.

“We were quite surprised to find that HITS, a query-dependent feature, is about as effective as web page in-degree, the most simpleminded query-independent link-based feature.”

in HITS on the Web: How does it Compare?
http://research.microsoft.com/research/pubs/view.aspx?0rc=p&type=Publication&id=1734

Great blog! Thanks for your posts.

]]>