Current Research in Digital History


Shawn Martin

Topic Modeling and Textual Analysis of American Scientific Journals, 1818–1922

from volume 2 (2019), https://doi.org/10.31835/crdh.2019.11

Abstract

When performing a distant reading of some of the most prominent American scientific publications in the nineteenth-century U.S., some very clear patterns emerge. LDA topic modeling and textual analysis methods of over one hundred years of the American Journal of Science (AJS), Proceedings of the American Association for the Advancement of Science (PAAAS), and the Journal of the American Chemical Society (JACS) between 1818 and 1922 helps historians to understand how these journals evolved over the nineteenth-century. Overall, there was an increase in discussion of business and professional issues and a shift in the journals that scientists used to discuss these issues. This shift happened during a very specific period, 1870-1890, the very same time that specialized scientific societies, particularly the American Chemical Society, split from the more generalized American Association for the Advancement of Science. Further analyses of these datasets may help to better understand shifts American science in the nineteenth century, and topic modeling methods allow historians of science to better identify this evolution of American science.

Introduction

How did science, as reflected in contemporary journals, change in the U.S. during the nineteenth century? In 1818 The American Journal of Science (AJS) was first American scientific journal published. The Proceedings of the American Association for the Advancement of Science (PAAAS) was published between 1848 and 1914 and also served as a news source, especially for news of the association. The Journal of the American Chemical Society (JACS) was first published in 1879, but it was not until the 1890s that JACS began to serve as a news source for the American Chemical Society and a space for discussion of theory and method in chemistry.

The differences between the content of these three journals can help both historians of science and academic publishing how American scientific journals evolved from sources of news to providers of original research over the course of the nineteenth century. Using general topic modeling to analyze the content of the major American scientific journals in the nineteenth century reflect some general scientific trends. In the mid-1890s, chemistry separated as a discipline from the more generalized American Association for the Advancement of Science, and consequently decreased as a topic within the more generalized scientific journals. In all, this shift in topics over time serve as evidence about American science in the nineteenth century.

Historiography

The changes in these journals over the course of the nineteenth century reflect many of the developments noted by researchers studying journal publication. Alex Csiszar has focused on scientific publishing in Britain and France during the nineteenth century, and Melinda Baldwin has concentrated on a case study of one particular journal, Nature, that started in 1869. According to Csiszar, British scientific journals attempted to address “the failure of the authority of the collectives that had traditionally adjudicated the boundaries of scientific authority.”1 Csiszar also argues that during the nineteenth century, there was a greater interest among scientists to create a sense of objectivity in the articles that they were writing.2 Furthermore Csisizar argues that expert communities of authors were becoming specialized.3 Melinda Baldwin’s work complements Csiszar’s points about the increasing demand by audiences for scientific content. Over the course of the century, Nature became a specialized scientific journal similar to what Csiszar generally describes.4

In the United States, the same trends hold, though in different ways. There was only one major scientific journal published consistently throughout the nineteenth century in the U.S. According to Simon Baatz, the American Journal of Science, was largely a general science journal able to bring together divergent metropolitan groups in cities like New York, Philadelphia, and Boston in ways that more localized journals such as the Transactions of the American Philosophical Society were not.5

Methodology for Analyzing Journals

To perform an effective distant reading of these three journals (AJS, PAAAS, and JACS), I performed several steps. After obtaining electronic copies of individual years of the journal, I created full text for these files. These three journals contained 209 text files (one text file for every year the journal published until 1922) and about seventy-three million words. These text files contain the entirety of the dataset used for this paper. I then obtained the OCR of the individual years, and used Mallet, a topic modeling tool that analyzes words via a Latent Dirichlet Allocation (LDA) model.

Rather than performing an analysis of an entire year, initially, the goal was to separate out the articles, and to compare individual articles. Unfortunately, after repeated experiments, it became clear that the structure of the journal was too inconsistent to be able to separate out individual articles using computational methods, much of it would need to be done manually.

Simon Baatz’s article was valuable as a guide for creating categories of content in our study of AJS and PAAS. Baatz discusses some of the historical developments with the journal as well as some of the larger social issues that Benjamin Silliman faced when he was managing the journal.To obtain the content categories, topic models were created for every year individually and each topic model created a list of 40 topics (or lists of related keywords) for each year. These topics were assigned to 17 categories. These categories included disciplines such as astronomy or chemistry, genres of articles such as geographical and coastal surveys, and broader discussions on issues such as publishing, theory, and news. The “other” category represents either topics that appear infrequently, for instance there are some music articles that appear early in the journal’s run such as music; additionally “other” represents topics that could not accurately be identified. After creating these categories, the number of times each category appeared during a given year were counted. These categories were then grouped into spans of 25 years so that some of the math for word counts would even out and to allow both for more direct comparison and for a smoother graphical representation of change over time in the visualizations. These changes allowed for clearer visualizations of the dataset.

After obtaining initial topic models from Mallet, I categorized and visualized the topics in spreadsheets so that it was possible to create graphs for each of the three journals and to compare how the three journals differed and how they changed over time. I performed initial analysis with yearly topic models rather than with individual articles. This initial analysis proved useful because the models could test the general theory of how journals did (or did not become) more specialized during the nineteenth century.6

Categories Used for Comparison of Scientific Areas

Original Category New Category
Astronomy Other Science
Botany Other Science
Chemistry Chemistry
Engineering Other Science
Fossils Geology
Geology/Mineralogy Geology
Geography/Survey Geology
Geography/Coastal Geology
Mathematics Other Science
Medicine Other Science
Meteorology Other Science
News News
Other Other Science
Physics Physics
Publishing News
Theory/Method Theory
Zoology/Biology Other Science
Table 1. This table shows the categories into which I originally placed topics and the broader categories I created to more easily compare different areas of science. Topics of Fossils, Geology/Mineralogy, Geography/Survey, and Geography/Coastal were joined into a single "Geology" category. I also combined news and publishing into a single category of "news." Finally, I put together all of the other disciplines into the "other sciences" category. Chemistry remained as a single category as well as Theory/Method, and Physics.

AJS and PAAAS were more directly comparable. Both of these publications produced generalized scholarship and were not specialized. Therefore, the topic models from Mallet often mapped to disciplines such as geology or chemistry. It was more difficult to compare JACS directly to AJS or PAAAS. JACS was a specialized journal for chemistry. However, by creating a topic list of twenty topics for each year of JACS between 1879 and 1922, a historical analysis using Browne’s (1952) history of the American Chemical Society, was employed to organize the data; this volume is still the most comprehensive account of the history of the American Chemical Society for the early period (1879-1922) of the society. Browne’s history also contains a history of JACS and an analysis of the topics contained within the journal. Topics produced by Mallet were organized into “chemistry” topics, or topics that were mentioned in Browne’s section on the chemistry topics appearing in the JACS, and “other” topics, or topics that are largely discussing business and professional issues that are not discussed in Browne’s section regarding JACS. Chemistry topics are largely related to chemical experiments. Unexpected topics usually contain words like “president, election, or committee” and often refer to the business of the society, methodology of chemistry, or other professional issues.

Journal Comparisons

The topic models demonstrate some interesting trends that accord with what we know about these journals through other means. Figure 1 shows that the journal becomes more specialized in geology and this may be evidence of scientists of the increasing specialization of AJS and leaving the more generalized journal for more specialized journals as they are created in the later nineteenth century. Figure 2 shows similar trends for the PAAAS. PAAAS and AJS also seem to illustrate Baatz’s argument that American scientific journals in the nineteenth century were an important news source for scholars scattered throughout universities in the U.S. In the case of PAAAS, its purpose as a news source is much more pronounced and business of the society represents a very large proportion of topics throughout the publication of PAAAS. In Figure 3, the topics dedicated to professional issues double during the tenure of one particular editor after 1890, perhaps because there was no generalized journal (such as AJS or PAAAS) to discuss such issues.

Line graph charting the change over time in the number of topics in six categories: Geology, Chemistry, Physics, Theory, News, and Other Sciences.

Figure 1. This line graph shows the number of topics within categories and that geology topics increase over time, whereas other topics generally decrease. The graph also shows that until about 1871, “other sciences” were significantly higher than geology. Also, in 1871 “other sciences” decline precipitously and geological topics increase and overtake “other sciences.”

Graph showing the percentage of topics in the *Proceedings of the American Association for the Advancement of Science* between eighteen forty-eight and nineteen fourteen. Theory is especially prevalent.

Figure 2. This chart shows the percentage of topics over time in PAAS. This graph illustrates PAAAS’ publication of theory more clearly. Within PAAAS theory is never less than 10% and often represents more than 20% of topics within particular years.

Line graph showing the percentage of topics on business and theory over time in *Journal of the American Chemical Society* between eighteen nineteen and nineteen twenty-two.

Figure 3. This line graph shows the percentage of topics on business and theory over time in JACS. Such topics were consistently a small number of topics in any given year (less than 20% of all topics). Nonetheless, the unexpected topics doubled during the tenure of one particular editor in the years 1890–1892, from 10% to 20% of topics, and that increase stayed consistent until the end of the dataset in 1922.

Conclusion

American journals in the nineteenth century largely reflect professional trends within the disciplines they served. AJS was one of the most important means of distributing scientific information during much of the nineteenth century. AJS was not under the control of any society, but rather under the control of one of the most prominent scientists at the time, Benjamin Silliman. Silliman’s journal, as AJS was called, started out as a news source for many scientific practitioners around the United States; it also specialized in geology, the most popular scientific discipline at the time. Though Silliman’s journal published articles many disciplines like astronomy and physics, by the twentieth century it specialized entirely in geology. PAAAS, a journal that overlapped with AJS in its period of publication (until it ceased publication in 1914), is quite similar to AJS. PAAAS also published articles about many different disciplines but served even from its inception as a news distributor for the society. Unlike AJS, however, PAAAS became entirely devoted to news (mostly contact information about members of the society) by the twentieth century, whereas AJS became more dedicated to articles about geology and less interested in distribution of news.

JACS combined many of the characteristics of both AJS and PAAAS. JACS specialized in publishing about chemistry from its inception. News and professional topics were a rather small, but consistent, part of what JACS was publishing. Over time, however, it becomes clear that chemistry as a topic declined in both AJS and PAAAS. More importantly, this decline happens at exactly the same time ACS forms its own society separate from AAAS, perhaps suggesting that JACS became a more important source for discussion of chemistry topics. Additionally, topics covering professionalization and news increased at the same time. In all, one can begin to understand the evolution of scientific topics in this preliminary analysis of some of the United States’ most prominent nineteenth-century scientific journals.


Bibliography

Baatz, Simon. “Squinting at Silliman: Scientific periodicals in the early American republic, 1810-1833.” Isis 82, no. 2 (1991): 223–244. https://doi.org/10.1086/355726.

Baldwin, Melinda. Making Nature: The History of a Scientific Journal. Chicago: University of Chicago Press, 2015.

Browne, Charles Albert. A History of the American Chemical Society: Seventy-five Eventful Years. Easton, PA: Mack Printing Company, 1952.

Csiszar, Alex. “Objectivities in Print.” In Objectivity in Science: New Perspectives from Science and Technology Studies, edited by Flavia Padovani, Alan Richardson, and Jonathan Tsou, 145–172. New York, NY: Springer, 2015.

Csiszar, Alex. The Scientific Journal: Authorship and the Politics of Knowledge in the Nineteenth Century. Chicago: University of Chicago Press, 2018.

Csiszar, Alex. “Seriality and the search for order: Scientific print and its problems in the nineteenth century.” History of Science 48, no. 3–-4 (September 2010): 399–434. https://doi-org/10.1177/007327531004800306.

Gross, Alan G., Joseph E. Harmon, and Michael Reidy. Communicating Science: The Scientific Article from the 17th Century to the Present. Oxford: Oxford University Press, 2002.


Notes

  1. Csiszar, “Seriality,” 426. 

  2. Csiszar, “Objectivities in Print,” 165. 

  3. Csiszar, The Scientific, 8. 

  4. Baldwin, Making Nature, 21. 

  5. Baatz, “Squinting at Silliman,” 223. 

  6. Overall, these three journals combined contain 209 text files (one text file for every year the journal published until 1922) and about seventy-three million words. These text files contain the entirety of the dataset used for this dissertation. Two topic modeling analyses were run on this dataset. First, an initial topic model of forty topics utilizing Mallet was performed on each of the 209 text files (or years) and compared between the journals, these topic models were then categorized and visualized in spreadsheets so that it was possible to create visualizations for each of the three journals and to compare those visualizations to see how the three journals differed and how they changed over time. After performing this analysis, a second topic model was run on the entire corpus. For this topic model, each of the 209 text files were split into documents of 1,000 words and a 500 topic model was created to investigate the corpus as a whole. See Appendices for topic models. 


Appendices


Author

Shawn Martin, Dartmouth College Library, Shawn.J.Martin@Dartmouth.edu, ORC ID logo0000-0002-9128-5528