Google Scholar vs Native Search -Updated

Peter Jacso has created a new "polysearch" engine that looks at Google Scholar vs Native Search.

Preliminary tests have shown that Google Scholar often retrieves far fewer unique items than the native search engines  of the publishers. On the positive side, Google Scholar links to citing references if the document was cited by journals indexed in Google Scholar, and provides the immensely useful citedness score of the documents. When Google Scholar  has more 'hits' for a query, they often turn out to be duplicates and triplicates (not always displayed adjacently) with a separate hit for the TOC entry, the abstract, the PDF file and (if available) the HTML file. Although their URLs are slightly different, they take you to the same spot in the archive. These are redundant and confusing, well illustrated when searching the IoP archive in the collection above.

(Link boosted from LISNews.)

UPDATE: Anurag "Mr. Google Scholar" Acharya sent along an email to note a caveat or two about the side by side analysis, and has graciously agreed to let me reprint his thoughts here:

Just saw the note re: the comparison to Native search. This particular implementation assumes that papers from a publisher are available on a specific web site (eg, arjournals.annualreviews.org for Annual Reviews and interscience.wiley.com for Wiley).  This is often not the case. Eg: for Annual Reviews, papers can be found on many hosts in the annualreviews.org domain, for Wiley, papers can be found on doi.wiley.com.  As a result, the comparisons can be quite misleading. The simplest way to see this for yourself is to select Annual Reviews (I mention this since it is the easiest) and do any query.  In the Google Scholar search box, replace site:arjournals.annualreviews.org by site:annualreviews.org. And then compare. Here are some queries that I tried:

yrast
"brownian motion"
"prion protein"

Pick your own.

On his advice, I did several test searches that reran the Scholar searches in Annual Reviews by lopping off the arjournals prefix and lo and behold, a lot more results start showing up. Still not match for match, but much better than the side-by-side analysis would have you believe. Fascinating.