Philosopher - 4:29 pm on Dec 17, 2003 (gmt 0)

Superscript,

What your seeing with your two sites may not be a filter. It can be explained with the CIRCA technology quite easily. As the white paper explains, the ontology originally consisted of a set amount of data. Obviously, that ontology has grown, but it is still finite.

Also, Google's Adsense product was originally developed by Applied Semantics. As we know, adsense is an advertising model, so obviously the data in their ontology is going to be dramatically skewed towards commercial terms as you wouldn't base your advertising model on non-commercial terms. This is likely why your commercial site was hit and your astrophysics site was not. I would imagine their isn't much advertising done in that field (doing a search for "astrophysics" on Google shows only two adwords currently).

It is very likely that, over time, the ontology will continue to grow and the algo will be applied across the board, as the white paper seems to indicate it is fairly self-learning when fed enough data, but for now, it seems it's only being applied to a those terms it understands.

As to Sids conclusions, those were mine as well. In fact there is a statement in the white paper, that seems to concur with the thought that too much of the same phrase may be bad...

The notion of focus is roughly analogous to the specificity of a concept, in that more specific concepts tend to be strongly related to a small set of things, and is somewhat inversely proportional to frequency, since more frequent concepts are less useful for discriminating particular contexts.

The above statement seems to say exactly that. Repetition of a single phrase over and over means less to CIRCA than that phrase plus other related phrases and tokens dealing with the same concept being used.

This would explain why highly targeted or optimized pages seem to not be ranking as well as more general pages. It may also explain why directory pages are ranking so well, as the link descriptions would generally contain many different tokens & terms dealing with that general topic.