n-grams (such as “how” and “the”) would remain relatively stable from year to year. Alternatively, a linguistically informed approach to reducing the size of the search space could be done through part-of-speech (POS) tagging, such that only those n-grams identified as noun phrases would be added to the dataset. However, state-of-the-art POS taggers are only about 95% accurate, implying that many interesting n-grams could have been overlooked had we taken this approach. Nevertheless, POS-based n-gram identification remains an option, especially when the corpus to be analyzed is extremely large.

Finally, we constructed a Web-based system to enable us to query, graph, and explore our Communications n-gram database, plot and analyze multiple n-grams simultaneously, and combine related search terms into a single result. For example, the search phrase “cellphone+cellphones, smartphone+smartphones” would produce a graph containing two lines, one representing the combined frequencies of the terms “cellphone” and “cellphones” over time, the other representing the combined frequencies of the terms “smartphone” and “ smartphones” over time. To try out our Communications n-gram tool, see http:// www.invivo.co/ngrams/cacm.aspx.

findings

Though we cannot expect to identify all ways the computing field has evolved in a single article, we do aim to provide a point of embarkation for future research. Beginning with big-picture considerations, we are confident saying the structure and content of Communications evolved significantly from 2000 to 2010. An analysis of our metadata revealed several striking, large-scale structural changes from 2000 to 2010. Over that time, Communications published an average of 306 articles per year, each containing an average of about 2,400 words. However, these averages obscured underlying trends showing that both the number of articles published per year and the average length of each article grew significantly, especially in more recent years. These trends (see Figure 1) imply Communications was providing more value to its readers than in

if, in the aggregate,
Communications
reflects what
is happening in
computing, then
perhaps existing
industry standards
should be refined
to more closely
approximate
real-world practice.

it had previously, since more recent issues contained more articles and words than earlier issues.

Changing focus

Continuing our investigation, we next extracted the 15 terms that experienced the most growth or decline in popularity in Communications from 2000 to 2010 (see Table 1). We hope you find at least a few trends in the table that are unexpected or interesting; indeed finding them is a primary goal of large-scale data mining. For us, we noticed that several of the terms showing the most growth were related to science and technology, while several of the declining terms were related to business and management. But is this observation anecdotal or a broader pattern in Communications? To answer, and to show how n-gram analyses can be integrated with more traditional analytic techniques, we conducted an interaction analysis comparing the n-gram frequencies for terms related to business and management with those related to science and technology. We identified related terms using Thinkmap’s Visual Thesaurus software ( http://www.visual- thesaurus.com), which is specifically designed for this purpose. We then extracted n-gram frequencies for the resulting lists of related terms, using these values to conduct our interaction analysis (see Figure 2). As shown in the figure, the average frequency of business- and management-related terms declined steadily from 2000 to 2010, while science- and technology-related terms became more common. Our interaction analysis indicated that the observed disparity was highly significant (t[5052] = 2.834, p < 0.01), providing statistical evidence of Communications’ evolving identity.

Changes in style

The style of the writing in Communications also evolved from 2000 to 2010. Authors seemed to be transitioning from the traditional academic style of writing, adopting instead a less-formal, more personal voice. Evidence of this change can be seen in the increasing use of words that refer directly to an article’s author(s) (such as “I” +143% and “we” +137%) and in the increased frequency authors spoke