About the size of Google Scholar: playing the numbers
Stephen J Bensman
notsjb at LSU.EDU
Wed Sep 10 14:17:40 EDT 2014
Enrique and Emilio.
I read your working paper with great interest as it deals with the same topic on which we are doing research here at LSU. To tell you the honest truth, I had trouble with its basic premise, i.e., that Google Scholar (GS) has a given size. I do not think that it does, and, if it does, it is meaningless. The real problem is what is the size of documentary set that is relevant to the search query.
The WWW and PageRank (the Google search engine) operate within what can be called the power-law or Lotkaian domain. Informetric laws also operate within this domain. On top of that, PageRank operates on what is called the probability ranking principle, by which the probability of relevance exponentially decreases as the number of inlinks decreases, i.e. below a certain point you are dealing with gibberish manufactured by the search engine itself. Therefore, there is a need for left truncation and determination of what can be termed the x-min. Since we are dealing with the Lotkaian domain, the x-min marks the point where the asymptote or “tail” on the x-axis for the items begins.
We are dealing with Nobelists, and what we have found is that with PageRank the set of relevant documents is conterminous with the researcher’s h-index and the “tail” of his GS citations distribution. In other words—whether by serendipity or not—the h-index is an excellent estimate of the x-min of a GS citations distribution. Below that is what the Germans would call a “Trummerzone” or rubbish zone largely manufactured by the search engine itself. This conterminous-ness is a validation of both the h-index and Google Scholar. The relevance of the set is also proven by the fact that the extreme outliers on the right messing up the tail are usually works on the topics for which the Nobelist won the prize. Case closed.
Every field has its statistical problem. With medical research it is right truncation, for every patient has to die before the results are really known. With the WWW and scientometric research, it is left truncation.
If you are interested in how I view how Google Scholar works, you can read our working papers at the following URLs:
I hope to post another working paper there next week that will really clinch the point. But who knows? I may be wrong.
Stephen J Bensman, Ph.D.
Lousiana State University
Baton Rouge, LA 70803
From: ASIS&T Special Interest Group on Metrics [mailto:SIGMETRICS at LISTSERV.UTK.EDU] On Behalf Of Enrique Orduña
Sent: Wednesday, September 10, 2014 5:15 AM
To: SIGMETRICS at LISTSERV.UTK.EDU
Subject: [SIGMETRICS] About the size of Google Scholar: playing the numbers
Adminstrative info for SIGMETRICS (for example unsubscribe): http://web.utk.edu/~gwhitney/sigmetrics.html
The purpose of this mail is to present our latest working paper, deposited on July 24, 2014.
We propose the inextricable task of knowing the size of this huge black hole looks like Google Scholar (GS). Anyway, as the title of the document (
About the size of Google Scholar: playing the numbers), we have begun to make accounts and using 4 different empirical methods we estimate that the number of unique documents (different versions of a document are excluded) should not be less than 160 million (as of May 2014).
Regardless of this particular outcome, which is itself significant (especially when compared with other scientific databases, and that gives us key clues about the amount of scientific knowledge that can be searchable, found and accessed to on the web), even more exciting is the methodological challenge of this assumption. It has not only forced us to devise various techniques for measuring the size of this dark object that GS is, but
applying them we have shed light, again, on various inconsistencies, uncertainties and limitations of the search interface tools used by Google. In short, we have learned more about what Google Scholar does or does not, and we want to share it with you all.
This research comes at a good time. We are not only almost celebrating the 10th anniversary of GS but also hearing some voices (from somewhere in Europe…) finally relying on the use of Google Scholar for scientific evaluation.
Now, when empirical studies (http://googlescholardigest.blogspot.com.es/p/bibliography.html) demonstrate every day that Google Scholar and its derivatives
a) measure with similar credit to traditional bibliometric indicators,
b) are the most used products by scientists (http://www.nature.com/news/online-collaboration-scientists-and-the-social-network-1.15711),
c) have unfortunately ended up with the competition (Microsoft Academic Search is in an unexplained hibernation, http://googlescholardigest.blogspot.com.es/2014/04/empirical-evidences-microsoft-academic-search-dead.html)
seems that certain euphoria unleashed. We are pleased, better late than never…
However, without wanting to lower the aroused expectations, we emphasize that the problems of Google Scholar for scientific evaluation are not technical or methodological (coverage, reliability and validity of the measures, records filtering performance…). Seminal limitations are those related with:
a) the ease with which GS indicators can be manipulated
b) the transience of the results and measures (in many cases difficult to replicate stably),
c) the technological dependence on companies that develop tools that come and go on the consumer product market (http://ec3noticias.blogspot.com.es/2014/04/la-new-new-horizontes.html-bibliometrics).
Google Scholar enthusiasts are now welcome; meanwhile we will continue vigorously in which we already proposed several years ago: to reveal with “data”
and not mere opinions
, the bowels of Google Scholar, and to reveal at the same time their strengths and weaknesses. So, like the old serials published, we can only promise...TO BE CONTINUED…
Polytechnic University of Valencia
Emilio Delgado López-Cózar
Universidad de Granada
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SIGMETRICS