nandeshwar.info » visualization

Mining publication data

a7n9 — Mon, 08 Mar 2010 15:57:24 +0000

I found treasure! Publication and citation data with metadata (author names, addresses, affiliation): http://citeseer.ist.psu.edu/oai.html

I was reading about knowledge management here, which says that knowledge management is nonsense. I agree to a certain degree, not because of the field, but because of its name. How do you manage knowledge? Isn’t knowledge derived? Wasn’t information “science” good enough? (I have problem with “business intelligence” as well…) As the author of that article says, it is a new term coined to attract attention. He does provide some evidence, but I was left unsatisfied.

I thought of performing text mining on publications database, and citeseer has this great resource. I downloaded the data (72 XML files), performed some clean-up, and ran a script to pull citeseer ID, author addresses, and publication dates where the abstract contained the term “knowledge management”. I was interested in seeing the trend of publication and places of publication.

Have a look at this chart:

There is a definite growth in this area, at least in research and publications. It is startling to see a paper published in 1970, and a peak in 2002. As citeseer data ends in 2004, it is possible that it doesn’t have complete publication history of 2004.

Geographic location wise, the US and Europe leads the way in number of publications:

My experiments with sparklines

a7n9 — Wed, 23 Sep 2009 18:05:19 +0000

For the uninitiated, Sparklines are “data-intense, design-simple, word-sized graphics” according to its inventor Edward Tufte. I always wanted to include them in my trend reports. The challenge: How and Which tool to use?

The data came from this report came from a complex query using SQL server and Access that had this format:

College	Major	2005	2006	2007	2008	2009
A	B	10	15	14	18	25

Some options I investigated and tried (be sure to read this page, you might find many more):

Google’s Chart API (http://code.google.com/apis/chart/types.html): you can embed these in Google spreadsheets, create HTML pages using Excel via VBA, or embed them in Excel sheets. Alas, none would work the way I would like them to work: In-cell graphics
Use Sparklines for Excel add-in: this add-in will create great in-cell charts (bar, bullet graphs, sparklines, etc), but copying them down is difficult and resource intensive, and any change you make in the column size will alter the shape of that chart. In addition, I had more than 400 rows to populate–these would be too many objects in a spreadsheet for Excel to handle. My machine froze when I tried to copy it for 10 rows.
Use R’s implementation by Jason Dieterle (search on this page for Jason Dieterle (email), January 28, 2008. Works very good. I modified the code to print max and mins only, but the function generates a graphic, which needed to be embeded in LaTex file. It did not work nicely. Charts were too big to fit in a cell of a table.
Create bar charts in Excel using REPT function. I tried different font sizes and styles, but it didn’t look pretty, for the range of data varied. (Hint: make the alignment of text 90, use pipe signs, create columns and graphs for each data value, remove gridlines, keep the columns very close. It almost worked.)
Use spark package for LaTex, doesn’t work in pdflatex, and you have to play a lot with the settings
Use sparklines package for LaTex. Problem was that data needed to be normalized (or scaled) from 0 to 1, and needed extra parameters for min and max points. Solution: create a normalize function in Excel, and write a big formula to produce the exact needed string for the sparklines to work i.e.:
\begin{sparkline}{5} \sparkdot 1 1 blue \sparkdot 0.2 0 red \spark 0.2 0 0.4 0.0625 0.6 0.5625 0.8 0.75 1 1 / \end{sparkline}

Here’s the normalize function:

Public Function Normalize(cell2Normalize As Range, WholeRng As Range)
       Normalize = (cell2Normalize.Value - WorksheetFunction.Min(WholeRng)) / (WorksheetFunction.Max(WholeRng) - WorksheetFunction.Min(WholeRng))
End Function

Here’s the big formula to convert the range for data values from 2005 to 2009 (in the range E2:I2) to the sparkline LaTex environment.
="\begin{sparkline}{5} " & "\sparkdot " & CHOOSE(MATCH(MAX(E2:I2),E2:I2,0),0.2,0.4,0.6,0.8,1) & " " & 1 & " blue " & "\sparkdot " & CHOOSE(MATCH(MIN(E2:I2),E2:I2,0),0.2,0.4,0.6,0.8,1) & " " & 0 & " red " & " \spark 0.2 " & Normalize(E2,E2:I2) & " 0.4 " & Normalize(F2,E2:I2) & " 0.6 " & Normalize(G2,E2:I2) & " 0.8 " & Normalize(H2,E2:I2) & " 1 " & Normalize(I2,E2:I2) & " / \end{sparkline}"

This last option worked beautifully. I dragged the formula down. I selected the data, and clicked on “Convert Table to LaTex” button (using this add-in). Copied the LaTex code to clipboard and pasted it in my LaTex editor. Manually merged the rows for colleges (using \multirow), and generated a beautiful looking pdf.

I was very happy. Printed it in color. Got a request back very soon that there should be total rows.

I forgot about the sparklines and created a report in Access with plain old numbers in less than 15 mins. Gave it back.

I tried to repeat this – Sparkline in Cognos -in Access with no luck.

(I later tried it one more time: Got the data in Excel using External data> Access, created a pivottable with rows and all, did some formatting, copied and pasted values and formats, inserted sparkline code, converted it to LaTex, copied and pasted in LaTex editor, and here’s the beautiful looking sample pdf of 16)

I wish there were simple reporting solutions that included awesome data visualization tools. (BTW, Excel 2010 will have sparklines: link) For this report, I did try Sweave, R, and LaTex, but because of the time constraints I could not investigate it further.

Please comment if you know any other way which meet (or don’t, Tableau is certainly one) these conditions: inexpensive (read free), efficient, and repeatable.

Data-Information Visualization

a7n9 — Thu, 21 May 2009 12:34:56 +0000

Ever since I read the book The Visual Display of Quantitative Information, by Edward R. Tufte, I am captivated by the idea of creating good design while doing data analysis or dashboard building. Although Excel 2007 charts are much nicer than its previous births, I have started disliking Excel charts. I am even developing an eye for picking out the bad information pixels. Apart from Tufte’s books, these books have helped me immensely:

The Elements of Graphing Data, by William S. Cleveland
Information Dashboard Design, by Stephen Few

Administrators/executives neither have the time nor the patience to understand complicated data mining algorithms and its results, and when they don’t understand them most probably they will never go in “production.” Simple, yet informative, designs and charts have better chances of going in production, which I am sure every data miner longs for.

I found a course web-site on Information Visualization: http://www.stat.auckland.ac.nz/~ihaka/120/lectures.html

nandeshwar.info » visualization

Mining publication data

My experiments with sparklines

Tag Cloud of Data Mining Jobs

Data-Information Visualization