For those of you who checked out gps31's msn white paper link, I just wanted to raise a few questions concerning the research done there. They talk about using VIPS to cut up a website into blobks and using semantic analysis to decide which blocks hold the most weight, then using the blocks to link to a page, and then the same happens again. The main flaw is that this strategy remains dependant on the assumption that pages will generally keep the same layout, but what about flash sites for example, will they not be taken into consideration? A spider and ranking module work together pulling apart pages and seperating semantic information from the overall meaning of the page. Is this wise?? Using semantic information to this extent would mean focusing on the wording of the links in relation to each othe, adding another dimension to seo, however this is far too easily taken advantage of. When compared to HITS and PageRank, the BLHITS seemed to return results closer to the "correct" or "expected" set. Who defined this "correct" set, or how was it calculated? Surely this would depend on what the user wants rather than what is "expected". I also noticed that they refer to the random surfer model once but use the concept many times in their calculations. The whole point of the random surfer model is that it is a predicted, assumed value, which in tests does add an adequate variable to the calculation, but isn't ever exact. A better way of doing this entire thing is to help the user better define what s/he is looking for - this method can't realy be flawed can it? It struck me that they assumed a lot of user behaviour and didn't back any of it up with figures or stats...where is the research? I agree that neural networks are strong classifiers and can be used to learn which part of a page is the most topical, but again, it could be manipulated and force us all into a set template.

I think that using blocks for semantic analysis and then finding relevant pages to the block is a nice idea, but I am not convinced this works in practice. I am most perplexed as to where the original ranking comes from also.

*sigh* there you go, just my endles blanter filled with questions...but at the end of the day we all know msn will do what they want. Lets just hope all creativity is not hampered by seo rules.

__________________
Programmer, researcher, ...

Search Science