Communications - May 2012

be headed. After all, “We are what we write, we are what we read, and we are what we make of what we read.” 4

Here, we address the identity prob-
lem by prescribing the same medicine
for ourselves—technology and algo-
rithms—we often prescribe for others.
We present a culturomic analysis5 of
Communications showing how natural
language processing can be used to
quantitatively explore the identity and
culture of an institution over time, in-
spired by the n-gram project released
in 2010 by Google labs (http://ngrams.
googlelabs.com). In natural language
processing, an n-gram can be viewed
as a sequence of words of length n
extracted from a larger sequence of
words. 12 Google’s project allows for
quantitative study of cultural trends
based on combinations of words and
phrases (n-grams) appearing in a cor-
pus of more than five million books
published as early as the 15th century.
The central theory behind the project
is that when taken together, the words
appearing in a collection of books re-
veal something about human culture
at the time the books were written.
Analyzing these words computation-
ally makes it possible to study cultural
evolution over time.

figure 1. structural changes in Communications from 2000 to 2010.

450 total words published per year articles published per year

1, 100,000

Words Published

900,000

700,000

400

350

300

Articles Published

500,000

2000

2002

2004

2005

2006

250 2010

table 1. major trends (growth and decline) in terms published in Communications from 2000 to 2010.

Growing in Popularity
Change
+ 18,151%
+ 11,160%
+ 10,833%
+ 6,439%
+ 5,703%
+ 5,151%
+ 5,151%
+ 4,844%
+ 4,599%
+ 4,292%
+ 4,231%
+ 4,016%
+ 3,741%
+ 3,557%
+ 3,434%

term Google queue cloud vM i T professionals parity workload venue polynomial time DRAM test cases theorem ooP science and engineering emulator

Declining in Popularity
Change
- 9,459%
- 9,459%
- 9,295%
- 9,295%
- 8,969%
- 7,991%
- 7,665%
- 7,502%
- 6,425%
- 6,197%
- 5,382%
- 4,281%
- 4,240%
- 4,158%
- 4,077%

term
perceptual
wrapper
biometrics
CoRBA
telemedicine
disintermediation
multimedia
transcription
personalization
user profile
e-commerce
e-business
satellites
AoL
oCR

method

To appreciate how the identity of Communications has evolved, we first constructed a corpus of the complete text of every article it published from 2000 to 2010.a We also collected metadata for all these articles, including title, author(s), year published, volume, and issue. In total, our corpus contained 3,367 articles comprising more than 8. 1 million words. To put this in perspective, consider that if you were to spend 40 hours per week reading Communications, you would need more than four months to read every article published from 2000 to 2010.

With our corpus complete, we next constructed a software system to tokenize, or split the text of each article into a series of n-grams. For example, René Descartes’ famous phrase “ cogito ergo sum” 10 can be subdivided into three 1-grams (cogito, ergo, and sum), two 2-grams (cogito ergo, and ergo sum), and one 3-gram (cogito ergo sum). As this illustrates, the number of n-grams that could potentially be extracted from a large corpus of text greatly exceeds the number of words in the corpus itself. This situation has serious scaling and performance implications for a corpus with millions of words, so to avoid them we limited our analysis to include n-grams with a maximum length of n = 4.

To address the challenges of punctuation, we adopted the same method used by the developers of Google’s n-gram project for digitized books, 13 treating most punctuation marks as separate words during the n-gram construction process. The phrase “ Elementary, my dear Watson” would be tokenized as, say, five words: