Chapter 3. Visualizations of a Digital Collection’s Data

This chapter presents an extended example of visualizing the digital collection of papers of the astronomer Clyde W. Tombaugh (1906–1997) at New Mexico State University Library Archives and Special Collections. A series of R plots offers insight into Tombaugh’s professional and personal life as revealed through his papers, as well as into the administrative side of managing the collection.

About Clyde W. Tombaugh

Clyde William Tombaugh was an American planetary astronomer who discovered the ninth planet in our solar system, Pluto, when he was twenty-four years old. His story is quite incredible. He was a farm boy in Kansas who liked to join his father and uncle when they stargazed in their spare time. Both men readily shared with Clyde what they learned during their informal studies and practice. The boy got inspired and constructed several telescopes with parts of discarded farm machinery. He began exploring the night skies on his own. Young Clyde was fascinated with Mars and Jupiter. When he sent his sketches of these planets to the Lowell Observatory at Flagstaff, Arizona, in 1929 asking for some feedback, its director V. M. Slipher offered him a job as a technician.

At the time, a new telescope at the Lowell Observatory had just been completed, and Tombaugh was hired to renew the photographic search for trans-Neptunian planet X. The mysterious planet X had been postulated by Percival Lowell, the founder of the observatory, in 1905. Tombaugh found planet X on February 18, 1930, a year after he was hired. He was the only American astronomer who found a planet in our solar system.1 The discovery was announced almost a month later, on March 13—Percival Lowell’s birthday and the date of the discovery of Uranus by William Herschel in 1781. In recognition of his spectacular achievement in observational astronomy, Tombaugh was awarded the Jackson Gwilt Medal and Gift by the British Royal Astronomical Society.2 In February and March 2020, the Lowell Observatory, along with New Mexico State University, celebrated the ninetieth anniversary of this incredible discovery by a young man who reached the stars.3

For Tombaugh, the discovery of Pluto was just the beginning of an exciting career. Awarded the Edwin Emory Slosson Scholarship to the University of Kansas, he pursued his education and earned bachelor’s and master’s degrees in astronomy in 1936 and 1939 respectively. In the meantime, he was still engaged in research projects at the Lowell Observatory. While working in Flagstaff, he discovered multiple clusters and superclusters of galaxies, clusters of stars, asteroids, and comets. He also fully explored what he called the Great Perseus-Andromeda Stratum of Extra-galactic Nebulae.4

During World War II, Tombaugh taught physics at Arizona State Teachers College and then navigation for the V-12 Navy program. In the following academic year, 1945–46, he worked for the astronomy department at the University of California at Los Angeles (UCLA), teaching both astronomy and the history of astronomy. He then moved to Las Cruces, New Mexico, where he was hired as chief of the optical measurements section at the ballistic research laboratory at the White Sands Proving Ground, and then as an optical physicist at its systems engineering branch. Tombaugh’s work there focused on German V-2 rockets, obtained at the end of war, and included designing tracking telescopes used to photograph missiles and rockets.

Involved in rocketry research during the Cold War and the space race, Tombaugh also initiated extensive satellite searches, originally conducted at the Lowell Observatory, then moved to the White Sands Proving Ground, and eventually transferred to the Physical Science Laboratory at New Mexico State University (NMSU) in 1955. The satellite search project led Tombaugh’s group to the Quito Observatory in Ecuador, where the photographs of the Russian Sputnik I were taken. No other small natural, near-Earth satellites were discovered. Tombaugh’s final report for NASA provided solid evidence for carrying out safe space exploration.

Tombaugh also initiated a photographic Planetary Patrol and Study Project of five planets: Mercury, Venus, Mars, Jupiter, and Saturn. The patrol was conducted from 1958 to 1973, mainly at the NMSU Research Center, where Tombaugh began working as an associate research professor in 1959. A year later, in recognition of his contributions to space sciences, he was awarded an honorary doctorate by Northern Arizona University. While at NMSU, Tombaugh returned to his earliest fascination and focused his research on Mars and its geology.

Tombaugh (figure 3.1) was one of the founding members of the department of astronomy at NMSU, where he taught until his retirement in 1973. During his tenure he also contributed to the establishment of the planetary observatory on the campus and the graduate program in astronomy.

In 1980, the fiftieth anniversary of the Pluto discovery, Tombaugh published his discovery account entitled Out of Darkness: The Planet Pluto, which he coauthored with Patrick Moore. The same year, he was inducted into the International Space Hall of Fame, established in 1976 at the New Mexico Museum of Space History in recognition of those who advanced human knowledge of the universe.5 In 1986, the Clyde Tombaugh Scholars Fund was established to support young space scientists in the department of astronomy at NMSU.6

Tombaugh conducted astronomical research and kept using his own telescopes until the end of his life. He used to comment on his career that he really had a tour of the heavens.7 Tombaugh died in 1997 in Mesilla Park, near Las Cruces, New Mexico. His ashes were placed inside New Horizons, an interplanetary space probe launched by NASA in 2006 to explore Jupiter, Pluto, and objects in the Kuiper Belt. The Tombaugh Regio, a large heart-shaped light-colored surface feature of Pluto, is named after him, as is the asteroid 1604 Tombaugh.

Clyde W. Tombaugh Papers Archival Collection

After Tombaugh’s death, most of his papers were donated to the New Mexico State University Library Archives and Special Collections for preservation and access. One hundred thirty linear feet of materials without any original order were structured and organized into a manuscript collection added to the Rio Grande Historical Collections.8

Initially, Tombaugh’s collection was divided broadly into personal and professional papers. These two categories were further organized into specific subseries, including extensive correspondence arranged either alphabetically by correspondent name or chronologically. Personal papers include documents related to Tombaugh’s youth in Kansas and his early interest in optics, telescopes, and stargazing, as well as his education and first work experiences at the Lowell Observatory. This part of the collection also covers his family correspondence, his deep commitment to the Unitarian Universalist Church, celebrations of Pluto discovery anniversaries, and fan mail. Tombaugh was a well-regarded figure in astronomy circles worldwide, and he also was highly popular with the public. He received tens of thousands of letters from amateur astronomers, aspiring scholars, and schoolchildren from across the world. He kept most of them.

Tombaugh’s professional papers were organized by the institutions he worked for: namely, the Lowell Observatory, Arizona State Teachers College, UCLA, the White Sands Proving Ground, and New Mexico State University. These papers include his voluminous correspondence with his fellow astronomers, astrophysicists, and other scientists and scholars from related fields, as well as his research reports and files, speeches and lectures, and writings. Several of his major projects are especially well documented in the collection, including the Near Earth Satellite Search, the Planetary Patrol and Study Project, and his lifelong studies on Mars, as well as his work on optics and construction of reflecting telescopes. The collection also includes firsthand accounts of his discovery of Pluto and subsequent research on it, which Tombaugh monitored closely over the years. The original photographic plates that Tombaugh used in his search for planet X, along with related documents, are held by the Lowell Observatory Archives in Flagstaff, Arizona.

In addition, the NMSU collection includes Tombaugh’s photographs, technical reports, maps, and design drawings of tracking telescopes. In 2003 Herbert A. Beebe, Tombaugh’s associate, donated to the collection a set of oral histories conducted with Tombaugh and his family, friends, and colleagues. These personal recollections of the past events, recorded in audio and video formats, added a final touch to this comprehensive collection.

The processing of the entire collection was completed in 2004. The finding aid, which thoroughly describes the scope, organization, and content of the collection, consists of 158 pages. This number directly speaks to the volume of the collection, which, in turn, reflects the productive life of a passion-driven scientist.

Clyde W. Tombaugh Papers Digital Collection

The digitization of Tombaugh’s papers started in 2015 with a goal of providing online access to his legacy. The structure of the digital collection reflects the arrangement of the physical collection, with specific series, boxes, and folders containing related documents. This mirroring of the archival collection was made possible by using features for collecting and storing compound objects provided by OCLC’s CONTENTdm platform. As a result, the organization and description of the digitized material are consistent with the description provided in the archival finding aid, which facilitates the work of researchers and students who use both collections.

Most of the series from the personal papers are already added to the online collection, as is a substantial part of the professional papers. In addition to text documents, the digital collection includes a number of photographs and a representative sample of various drawings that Tombaugh did of telescope designs, focusing devices, optical systems, satellites, planetary sketches, and constellation charts. A screenshot of the landing page of the Tombaugh digital collection is shown in figure 3.2.

Photographs at the top of the landing page, along with a short description of the collection, are meant to provide quick insight into collection content and its historical context. In addition to a brief overview of Tombaugh’s scientific career, the text introduces the status and progress made in the development of digital collection. Divided into two main categories, a list of links to various subseries allows a reader to get a sense of the materials included. It is impossible, however, to get an estimate of the collection’s breadth and volume based on this brief introduction.

Similarly, neither browsing through the collection nor reading the metadata describing folders and individual pages of documents provides a comprehensive overview of the collection. Due to limitations of screen displays, one can examine only a few records at a time while browsing or just focus on a document page (see figure 3.3).

Visualizations of the collection content can remedy these shortcomings. By providing overviews of a collection, visualizations help a reader to see not just one page, but all pages at once.

In the next section, eighteen charts provide holistic views of the collection and offer some additional insights into its content that both library users and cataloguers may find valuable.

Selected Visualizations Created for the Clyde W. Tombaugh Papers Digital Collection

What follows are visualizations that address specific questions about the Tombaugh Papers digital collection. Each plot is described and interpreted briefly. Likewise, a script for each plot is presented and explained in some detail. The R code for a given plot is divided according to the stages of a basic workflow, that is: loading R packages, loading data, transforming data, and plotting. In addition, in order to demonstrate what sets of values are translated into graphic forms, relevant data tables or their fragments are included.

Please note that the scripts provided below can be replicated and used to create visualizations with different sets of data. The readers of this report are encouraged to use the information below and start experimenting with R.

Question 1: How Many Documents Were Created in Successive Years?

Plot 1 (figure 3.4) is a time line graph. It gives an overview of the collection at a glance. This plot is the first holistic view of the Tombaugh collection. One can see at once how many documents of all categories were produced in successive years between 1900 and 2000. Apparently, the 1950s are best represented in the collection. This is the time when Tombaugh worked at the White Sands Proving Ground and at New Mexico State University.

In order to create a plot, first load the needed packages by calling the function library():

• library(data.table)—for data mining and processing
• library(ggplot2)—for data plotting
• library(stringr)—for string operations, such as detecting and extracting character patterns

Second, load the exported data into a data table called “data.” This is done by using the function fread() included in the data.table package:

Then change the original names of the columns. R does not accept column names with spaces. For this, use the function setnames(). Call the function setnames() with data and then for the original name, substitute a new name linking the words with an underscore:

Next, use the function setcolorder() to change the order of the columns to see the one of interest up front in the table:

In the next line, extract the year from the string date_original and store it in the new column “year” as a numeric value. This operation is performed on all rows in the “date_original” column.

The function str_extract(), which comes from the stringr library, is used here to extract a string representing year from the string in column “date_original.” Using regular expressions,9 this function looks for a pattern of four consecutive digits (using the expression “\\d{4}”). Then, the string representing the year is converted to a numeric value by another function, as.numeric(). Now, the numeric value gets assigned to a new column, called “year,” created (in place) in the table “data” with the operator ":=".

In the next step, produce annual counts of records for each year. Using the data.table syntax, create a new data table with two columns, “year” and “annual_count,” the latter one filled with the number of all rows (.N) for each year (by = year):

This operation is then followed by ordering rows with the function order(), which sorts annual counts chronologically. All the records with missing dates are deselected by the function na.omit(). The result of all operations is then stored in the data table “annual_counts,” which has two columns, “year” and “annual_count,” as shown in table 3.1.

To produce a plot, call the ggplot() function with the data table “annual_counts” as data, and then within the same function define the x and y axes with the aes() function. For x, use the column “year,” and for y, use the column “annual_count” defined before in the data table “annual_count_original.”

In the next line, define the plot type. The geom_col() function stands for the bar plot with arguments color and fill specified by “deeppink4”. Note that the plus signs mean adding additional layers to the plot. In the next line, define the range of years for the x axis (1900 to 2000), and set tick spacing (breaks) to every ten years. Then, define labels for the x and y axes labels as “Year” and “Annual Count of Pages” respectively.

In the next step, define the title for the entire plot as “All Documents.” The last line calls the function theme(), which sets the central position and the font size of the plot title. The theme() function can be used to define many other details of the plot, such as font sizes of axis text, axis tick intervals, and so on.

Question 2: How Do the Volumes of Personal and Professional Correspondence Compare?

Plot 2 (figure 3.5), a bar plot, also displays an overview of the collection at a glance. Bar plots are employed to show the relationship between numeric and categorical variables. This bar plot compares the volumes of two categories, personal and professional papers. At the current stage of collection development, personal correspondence dominates by almost twofold.

After loading packages and data as before (see script for plot 1), create a new column, “series_category,” by extracting all characters before the comma (using the expression ,.*) from the original metadata column “Series,” which also includes box and folder numbers (e.g., extracting “General Correspondence” from “General Correspondence, Box 004, Folder 001”):

In the next step, create two lists of series categories, series_personal and series_professional, as defined in the finding aid (see the transforming data section of the script for plot 1).

Then follow by creating another column, “per_prof,” and assigning to it either the name “Personal papers” or the name “Professional papers,” depending on which of the two defined lists the document series category belongs to:

In the next line, count documents for each of the two lists, personal versus professional, create a column to store the counting results, and then remove rows with missing values. The result of these operations is then assigned to the data table “per_prof_counts,” as shown in table 3.2:

The final step involves creating the bar plot by calling the ggplot() function.

Use the newly created data table “per_prof _counts” as input data. Then define aesthetics, including the axes and the bar color and fill that indicates the category—personal or professional. The function scale_fill_manual() allows you to customize colors selected for the bars. In the next line, define the plot type. The geom_col() function creates the bar plot. This time no arguments are used. The labs() function defines the axis labels. In this plot, there is no label for the x axis. Instead, the category names appear below the bars. With function theme(), you define the font size (size = 9) and the centered location (hjust=0.5) of the tick labels (“Personal papers” and “Professional papers”). In the last line, by setting legend.title to element_blank(), you remove the title from the legend, which in this case would not add any information.

Question 3: How Many Documents in Each Category Were Created in Successive Years?

Plot 3 (figure 3.6) compares two time line graphs presented together. This chart provides insight into when personal and professional documents originated.

The comparison between the time lines shows that Tombaugh was exchanging personal letters pretty much his entire adult life. His professional correspondence time line reflects his engagements with five institutions. Evidently, his professional activities in the 1940s through the 1960s, when Tombaugh was working at Arizona State Teachers College, UCLA, the White Sands Proving Ground, and New Mexico State University, are best represented in the collection to date. The peak in 1929–30 in the personal papers category corresponds to Tombaugh’s first year at Lowell Observatory and the discovery of Pluto, whereas the peak in 1980, again in the personal papers category, corresponds to the fiftieth anniversary of this discovery. Both events are well documented in the digital collection.

An additional step in transforming data and getting them ready for plotting is creating annual counts of document pages by personal and professional categories, stored in the data table called “per_prof_counts_by_year.” A fragment of this data table is shown in table 3.3.

Next, call the ggplot() function again. A new element in this section is the facet_wrap() function, which lays out two plots in one figure:

The tilde before per_prof makes the plots split by the values in the column “per_prof” (“Personal papers” and “Professional papers”). The parameter ncol=1 lays out the plots in one column. The parameter scales=”fixed” sets the same axis scale for both plots. The function theme() defines font sizes for titles in subplots, as well as font sizes and colors for axis x tick labels, axis titles, and legend labels. Again, the title for the legend is removed.

Question 4: What Are Document Volumes in Selected Series?

Plot 4a (figure 3.7) is a horizontal bar plot with volumes per specific series. On this graph we zoom in to see personal papers in a greater detail. The plot displays counts for all series, which allows readers to assess and compare their values. Visibly, the General Correspondence series holds the most items.

To produce series counts in personal papers, rows in data are filtered with the following statement:

Extract only those rows for which the series cate-gory value is in the list “series_personal” (see the script for plot 2 for the definition of the list “series_ personal”). Table 3.4 shows the pages counts for series in personal papers.

In plotting, there are four new function calls here: reorder(), coord_flip(), geom_bar(), and ggtitle(). With the function reorder(), reorder the category levels by their counts. The coord_flip() function swaps x and y axes. Axis y maps the category count, and axis x maps category levels for column “series_category.” Then, the call to the function geom_bar() with stat=’identity’ ensures that the heights of the bars correspond to the series counts. In general, the parameter stat defines the function on aggregated data (by default, ‘count’). You counted the records previously, so you want geom_bar() to take the counts directly, without aggregation. Hence, you use stat=’identity’. The argument fill defines the fill color of the horizontal bars.

Next, using the function labs(), define the labels for the flipped x and y axes, and with the function ggtitle(), the title of the entire plot. The last function, theme(), helps defining the center position (hjust=0.5) and font size of the plot’s title.

Plot 4b (figure 3.8) is another horizontal bar plot with volumes per series. It shows the distribution of series counts for professional papers. Here, the NMSU Research Files series holds the most records to date.

The same R code functions apply here. The only difference from plot 4a is that now we take into account only the column “series_professional” as displayed in table 3.5.

Question 5: When Were the Documents in Each Series Created? Is There Any Time Overlap?

Plot 5a (figure 3.9) is an aggregation of several time lines for selected series for comparison.

On these graphs, we can see the distributions of series in personal papers over the years. Again, we see corresponding peaks around 1980 in the Awards and Honors series and the Pluto Discovery Anniversary series. It was the fiftieth anniversary of the Pluto discovery event, “the Golden Year of the Ninth Planet”, that brought so much recognition to Tombaugh.

Interestingly, Fan Mail’s peak seems to correspond to the same event. Presently only the first few years of fan mail are included in our digital collection. But based on what we can see here, it looks like Tombaugh was not popular immediately after making his discovery. It was more toward the 1950s when his publicity really grew. The Planetary Observation series represents his early interests in astronomy, while the Telescopes and Optics series shows his lifelong engagement in constructing reflecting telescopes.

Here repeat the first steps from the previous plots, starting with plot 2. In the transforming data section, select a subset of series to include in the plot and assign the list to “series_personal.” Then calculate annual counts for all series and title the resulting data table “annual_counts_series.”

Using the function ggplot(), create multiple time lines in one plot. First, assign the data table “annual_count_series” to data with its rows filtered (with “series_category” values belonging to the “series_personal” list). A fragment of the “annual_counts_series” data table is shown in table 3.6.

In the next step, map the x axis to “year” and y axis to “count” in the aesthetics function (aes()). Then call the function geom_col() to create bar charts and choose the fill color for bars. With the function xlim(), define the range of years on the x axis. Next, define the plot title (with the function ggtitle()) and set its position to centered with the function theme() (horizontal adjustment, hjust=0.5). In the line, define axis labels with the function labs().

Finally, call the facet_wrap() function, which arranges multiple plots (one for each level of “series_category”) in three columns (ncol=3). Also, the y axis scaling is adjusted for each plot (scales = “free_y”), while x axis scaling stays the same. This is done to accommodate the different numbers of records in each series.

Plot 5b (figure 3.10) shows multiple time lines for professional papers for comparison.

Among several institutions that Tombaugh worked for, the correspondence associated with the Lowell Observatory, the White Sands Proving Ground, and New Mexico State University, including NMSU Physical Science Lab, are well represented in the digital collection. Papers produced at the Arizona State Teachers College and UCLA are of more limited scope.

The same script is used here, only with a different subset of data and custom ordering of series categories.

After creating annual counts for all series (“annual_counts_series”), filter them down to “series_professional” and assign the name “annual_counts_professional” to the resulting data table. With the function factor(), order the levels of the “series_category” column in “annual_counts_professional” as defined in the list assigned to parameter levels:

In the next step, implement reordering of “annual_counts_professional” by assigning the newly defined “ordered_factor” to the column “series_category.” Table 3.7 shows a fragment of the data set for plot 5b.

Finally, subject “annual_counts_professional” to the ggplot() function with the same parameters as in the plot 5a, keeping the same color (“deeppink4”) and changing just the title (“Professional series”).

Question 6: Who among Tombaugh’s Correspondents Was Most Prolific as Represented in the Collection?

Plot 6 (figure 3.11) is a horizontal bar chart with flipped coordinates. It compares volumes of letters written by different authors. It shows total counts of records per selected creator. As expected, Tombaugh, the author of a massive number of letters, is totally off the chart. Hence, there is no bar representing him at the top of the plot. If his correspondence were included, differences among other creators would not be well represented on the plot. Interestingly, based on the evidence in figure 3.11, Tombaugh corresponded most intensely with his students Brad Smith, Charles Capen, and Jimmy Robinson. His exchanges with his family members and colleagues come right after.

This chart also shows a problem with inconsistent format for the creators’ names. Brad Smith is listed here twice, as “Smith, Brad, 1931-” and “Smith, Brad, 1931-2018.” Apparently, records added more recently include an authority name updated after Smith’s death. Evidently, such inconsistency influences the outcome to a considerable extent. Visualizations provide a remarkable insight into quality of metadata and allow for quick identification of issues.

After loading needed packages and data, extract the column “Creator” from the original data as a list of strings and assign it to the column “creator_column”:

Unfortunately, some entries in the column “Creator” contain groups of creators (separated by semicolons), as shown in the example in table 3.8.

In order to calculate counts of pages for each creator correctly, first collect a long list of individual names of creators by splitting multiple names where needed, and then count occurrences of individual names. Collection of individual names is done in a for loop. First, create an empty list for individual names, “creator_list.” Next, set the loop to traverse all entries in the column “creator_column”:

In the for loop, execute the block of code surrounded by curly brackets. With the function strsplit(), split each column_entry with the separator “; “ as in the original data:

This operation results in a list of individual creator names (or just an individual name, if no separator is encountered), which is then appended to the list “creators_list,” using the function append():

The new list is assigned back to the list “creators_list.” After the loop ends, “creators_list” contains both individual names and, occasionally, lists of names. With the function unlist(), it is turned into a list of individual names:

In the next step, create the data table “creators,” with a single column “Creator,” holding elements of “creators_list”:

Then, from the newly made “creators,” produce counts of pages by “Creator” in descending order and store them in the data table “creator_counts” (fragment shown in table 3.9).

At this point, feed the first thirty rows of “creator_counts” (creator_counts[1:30, ]) as data to the ggplot() function. In the aesthetics function (aes()), axis x maps creator names (column “Creator”) ordered by “count,” and axis y maps “count.” Next, the x and y axes are swapped using the function coord_flip(), and a bar chart is created with the function geom_col(), colored in “deepink4”. Finally, x and y axis labels are created with the function labs(), and a range for axis y is set with the ylim() function.

Question 7: What Is the Distribution of Document Counts across Series for Individual Creators?

Plot 7 (figure 3.12) is a heat map. Heat maps show a color-coded view of multidimensional data. Here we look at three variables—creators, series, and counts—to see patterns of count distributions across series and creators. Missing values are not colored. This heat map provides a great way to identify who contributed to what correspondence series or, to put it differently, in what series one should look to find letters of a particular author. Apparently, some authors appear in different series, which may mean that they worked with Tombaugh on more than one project.

As before, start with loading needed packages and data. To create a heat map we need an additional package, called viridis, with color scales. This package needs to be installed before loading it.

Here, the data table “creator_counts,” created for plot 6, is used. As in the former plots, define “series_personal” and “series_professional” lists of “series_categories.” Next, create a new column, “series_category,” by extracting all characters before the comma from the original metadata column “Series,” which also included box and folder numbers (e.g., ”General Correspondence” from “General Correspondence, Box 004, Folder 001”):

Now, filter the rows of data down to series category values that are in either of the lists, personal or professional. Then, create the first segment of “counts_creators_series” containing counts of just Clyde Tombaugh’s papers per series category. For this purpose, filter rows down to those that contain the creator’s name. (We may encounter a group of creators’ names, not just an individual one.)

Then add an extra column, “Creator,” with the creator’s name. With the largest number of documents, Tombaugh is in the first position in the data table “creator_counts”:

Do the same with the selected creators in the for loop, where we concatenate vertically “counts_creators_series” with data segments corresponding to subsequent creators. A fragment of “counts_creator_series” is shown in table 3.10.

In the plotting section, define the heat map using the geom_tile() function with data as counts_creators_series and Creator and series_category as x and y axes, colored according to count values. Then define the color scale using the function scale_fill_viridis() with range of count values between 0 and 500 (limits = c(0, 500)), with truncating values outside of the range (oob = scales::squish). Finally, using the function theme(), define the angle of the x axis tick labels as 45 degrees.

Question 8: What Are the Most Dominant Terms in a Given Correspondence Series?

Plot 8a (figure 3.13) is a word cloud. A word cloud, known also as a tag cloud or a term cloud, is a visual representation of text data. This type of visualization is suitable for quickly perceiving the most frequent terms and, perhaps, getting at the essence of the text. The prominence of the relative count of each term is shown with font size or color.

The terms in this cloud were extracted from the Pluto Discovery Anniversary series in personal papers. As can be seen, the terms displayed in figure 3.13 correspond to the occasion. The word cloud in figure 3.14, representing a different series, displays clear difference in vocabulary used.

Plot 8b (figure 3.14) is a word cloud for the Arizona State Teachers College series. Plainly, Tombaugh taught navigation there.

In order to build a word cloud, first load the needed packages and data. Next, create a new column, “series_category,” as in the former plots, by extracting from values in the column “Series” strings of characters occurring up to the first comma.

Then select the column with OCR and preprocess it. The preprocessing of the “OCR” column involves turning its content in place into text format (with the function as.character()) in lower case (with the function tolower()) and removing all digits (with the function tm::removeNumbers()). Next, turn all non-alphanumeric characters into white space. For this operation, use the function str_replace_all() with a regular expression pattern “[^[:alnum:]]”. Then, remove excess white spaces with the function str_squish().

After preprocessing, select rows for a given series of interest (in data_subset) and turn them into a corpus (i.e., a set of text documents) by applying to the column “OCR” a sequence of functions, VectorSource() and Corpus(). The function VectorSource() produces a vector of text documents, each corresponding to a row in OCR column. Then, the function Corpus() turns the resulting vector of documents into a corpus of documents.

Next, use three function calls to remove English stop words from the corpus. The function tm_map() calls the function removeWords(), which deletes from the corpus the English stop words returned as a list by stopwords(“english”).

Then turn the corpus with all documents into a term-document matrix (“term_doc_matrix_tdm”) with terms in rows and documents (or records) in columns. Each entry in this matrix is the count of a given term in a given record. In the next step, turn the term-document matrix into an R matrix (“term_doc_matrix”), shown in table 3.11.

Some words displayed in the matrix may not make much sense because OCR may not be capable of rendering handwritten texts.

In the next step, sort the term-document R matrix by word counts across all records. The sorting outcome (a vector, “word_frequencies_vector”) is cast as a data table “word_frequencies” that has two columns, “word” and “freq,” shown in table 3.12.

We may want to add bigrams, such as “las cruces,” “white sands,” or “new mexico,” to the table “word_frequencies,” For each bigram, create a single-row data table with two columns, “word” and “freq.” Then assign the bigram to column “word” and its frequency to column “freq.” The frequency of the bigram is calculated as a sum of counts of its occurrences across all rows of the table “data_subset,” for example:

Using the function rbind(), append vertically all the bigram data tables to the table “word_frequencies” and sort it with the function order() into descending order by frequency.

After that, call the function wordcloud2() with two arguments: the data table “word_frequencies” limited to the top 150 entries, and shape. Different shapes for the cloud, including triangles, circles, pentagons, and diamonds, can be selected.

Question 9: How Often Were Particular Terms Used in Subsequent Years?

Plot 9 (figure 3.15) is a set of time histograms showing annual counts for selected terms used in Tombaugh’s correspondence over years. While terms such as home, money, and photography were frequently mentioned over Tombaugh’s lifetime, computer, missile, and satellites are more time-bound.

As before, load the packages and data, use the function setnames() to change the column name, and then use the function setcolorder() to place the column “date_original” at the leftmost position in the table for easier examining of the table. Then create a new column, “year,” by extracting four-digit year values from dates with the use of the function str_extract() and the regular expressions pattern (“\\d{4}”).

In the next step, select a list of terms for time line display. Then create an empty data table called “term_count_year” with columns “year,” “annual count,” and “term.” This table will store annual counts of all selected terms.

In a for loop, for each of the selected terms, create a two-column data table, “term_count,” with a column “count,” with counts of a given term in each document, and a column “year.” Based on this data table, calculate annual counts of the given term and store them in the data table “term_count_year_t.” Add to it another column, “term,” with the actual term. At this point, append the table “term_count_year_t” vertically to the table “term_count_year,” which stores annual counts for all the selected terms. Table 3.13 shows how it looks for the term photography.

In the plotting section, use the function ggplot() with data table “term_count_year” as data. Using the function aes(), define axes x and y, as “year” and “annual count” respectively. Then call the function geom_col() to draw a plot and define color fill for bars (“deeppink4”). Another function, facet_wrap(), places six histograms together on one plot. Each histogram corresponds to one of the selected terms. The argument ~ term splits data for each subplot according to values in the column “term.” Another argument, ncol, defines the number of columns in the arrangement of subplots, and the next argument, scales = “free_y”, means that each subplot has its own optimal vertical scale. Finally, with the function xlim(), define the range of values for the x axis.

Question 10: Where Did Fan Letters Come From?

Plot 10a (figure 3.16) is an interactive map of Tombaugh’s fans across the world. The dots represent locations from which Tombaugh received fan mail. Each dot corresponds to a letter. If a user hovers over or clicks on a dot, the corresponding address is displayed in a pop-up window. The digital collection holds just a representative sample of fan mail received by Tombaugh over the years.

Plot 10b (figure 3.17) is another interactive map, this time of the United States, showing locations of fan mail authors in a greater detail. Again, each dot represents a fan letter.

Before creating a script for the maps, set up an account with Google Maps to obtain the API key and get access to Google’s services.10 This step is necessary to produce geospatial coordinates (longitude and latitude) from addresses.

Start with loading packages and data. The data file has a column named “addresses” that holds all addresses extracted from the envelopes sent to Tombaugh. We use here only names of cities, states, and countries. We store the content of the file in the data table “fan_mail_locations.” In the next step, using the function register_google(), register your R session with Google Maps Platform with the previously obtained API key. Next, transform the data. Specifically, create a list of addresses (“addresses_list”) from the column “addresses” in the data table “fan_mail_locations.” Then, call the function geocode() with addresses_list as data to get the coordinates for each address. Store the results in the data table “geocode_addresses.” A fragment of this data table is shown in table 3.14.

Now the data are subjected to the function leaflet(), which creates a map widget. The next two function calls add layers to the map. The function addTiles() adds a tiled map. The function addCircleMarkers() adds circle markers to to “fan_map” at locations defined by the corresponding longitude and latitude values (coming from columns “lon” and “lat” in the data table “geocode_addresses”). The function addCircleMarkers() also defines markers’ radius, color, stroke, and opacity as well as two interactive features (popup for clicking on and label for hovering over markers) to display specific addresses associated with the markers. Finally, display the map with the function print().

To display the second map limited to the US (plot 10b), filter “geocode_addresses” to the ones in the specified range of longitudes and latitudes (see the last command in the script for plot 10) and then repeat the sequence of commands with the filtered addresses (“geocode_addresses_usa”), starting with the call to function leaflet().

Question 11: How Has the Collection Developed over Time?

Plot 11 (figure 3.18) is a set of three histograms stacked together. The histograms show annual counts of documents that were digitized, submitted to CONTENTdm (created), and modified, respectively. Based on the presented data, most of scanning work was done in 2015, whereas work on metadata and quality control intensified in 2018 and 2020.

To create this plot, we use three time-related columns, “Date digital,” “Date created,” and “Date modified,” from the data table “data.” “Date digital” refers to the date a document was scanned. “Date created” refers to the date a record was created, and “Date modified,” to the last date when the record was changed.

As usual, first load packages and data. Then follow the usual set of procedures to produce annual counts of time events for each type: digital, created, and modified. To do so, use the function setnames() to change the column names for the three columns, to “date_digital,” “date_created,” and “date_modified,” and then the function setcolorder() to place all three column in the leftmost position in the table, again for easier examining of data. Yet again, create a new column for each type by extracting four-digit year values from dates, with use of the function str_extract() and the regular expressions pattern “\\d{4}”. In the next step, create three tables with annual counts of time events for each type. Using the function na.omit(), remove all rows with missing values.

In order to produce a multifaceted plot with annual counts, take two additional steps. First, in each of three annual count tables, create a new column, “type,” storing the type of time event (“Digital,” “Created,” or “Modified”). You need the column “type” in order to later split the display into term-specific subplots. Next, stack the three annual count tables vertically (to produce the table “annual_counts”), but before doing that, change the names of the columns storing the year value for each type to the common column name “year.” To do this, use the function setnames(). In the next step, define the ordering in column “type” using the function factor() with argument level set to a list of types in the preferred chronological order (“Digitized,” “Created,” “Modified”). The complete data set used for plot 11 is shown in table 3.15.

For plotting, call the function ggplot() with the data table “annual_counts” as data. In the aesthetics layer, use the function aes() to define x and y axes as “year” and “annual_count” respectively. Next, with the function geom_col(), define the geometry of the plot as bar chart, with filling color “deeppink4”. Then use the function scale_x_continuous() to define the locations of tick marks, by setting the parameter breaks to a sequence between 2015 and 2020 (breaks = seq(2015, 2020)).

In the next step, call the function facet_wrap() to place the three histograms, defined by the values in column “type” (~ type) together on one plot. Arrange subplots in one column (ncol=1), and allow vertical scales to be optimized independently for each subplot (scales=”free_y”). Finally, define axis labels using the function labs().

Question 12: What Are Some Problems in the Metadata Fields?

Plot 12 (figure 3.19) is a horizontal bar chart for series counts displaying missing values and issues with series names, such as misplaced or absent commas. The chart demonstrates what the data we work with really look like and thus helps to identify recurrent problems in metadata descriptions. Based on visualizations like the one in figure 3.19, problems with messy metadata are quickly detected and fixed.

Next, calculate counts of records by unique values in the column “series_category” and store the results in the table “count_serie_category. Table 3.16 shows a fragment of data subjected to plotting.

At this point, feed the rows of “count_series_category” with values smaller than 120 pages ([count<120,]) as data to the ggplot() function. In the aesthetics function (aes()), map axis x to the column “series_category” (reordered by corresponding counts) and axis y to “count.” Next, by calling the function geom_bar(), define the geometry layer as bar chart, with filling color “deeppink4”.

Next, x and y axes are swapped using the function coord_flip(). Set the plot title with the function ggtitle(). Finally, with the function theme(), define the title’s center location (hjust=0.5) and font size (size=12).

R Shiny—Engaging Library Users in Producing Visualizations without Writing Code

R Shiny opens the analytical and visual power of the R programming language to users who do not have any background in coding. With an interactive user interface (UI), R Shiny allows users to formulate specific questions about digital collections and address those questions with visualizations.

There are two components of an R Shiny application: UI and server. UI, usually a web page, is composed of input and output panels. With input panels, including objects such as sliders, date ranges, drop-down menus, checkboxes, and so on, users can modify parameters that the R scripts will run with on the server. Output panels, such as plots, tables, text fields, and so on, display results produced by the server. The output is modified by the server any time the user modifies the input.

The server can run locally (as in the example presented below), but also can be set up as a web server and be available through the cloud. For a detailed introduction to R Shiny, please check the tutorial on the RStudio platform.11

An R Shiny app can start from the following template:12

In the fluidPage() function, set the page layout by assigning to its arguments different UI objects. While designing these UI objects, call input and output functions of choice to define the performance of the whole page. All analysis and visualizations take place in the server() function based on input. The results are assigned to output. Finally, in the shinyApp(), the user interface is connected with the server.

Plots 13a and 13b (figure 3.20 and 3.21) show interactive R Shiny versions of plots 4a and 4b that compare volumes of documents in selected series in personal and professional papers.

As before, first load packages and define the user interface. The arguments to the fluidPage() function, which sets the page layout, are tags$style(), titlePanel(), sidebarPanel(), and mainPanel(). Using tags$style(), define font sizes used in select list (selectInput()) and date range (dateRangeInput()) input controls. In titlePanel(), set the title of the page and its center location. In sidebarPanel(), define two input panels—for the select list (selectInput()) and the date range (dateRangeInput()). In both, set unique names for these inputs (inputId) by which they are identified in other parts of the app code, as well as their labels (label), which are displayed in the UI. These labels should help explain the functions of these panels. There are also input-specific arguments. For selectInput(), it is choices, defining a list of available values to select from. For dateRangeInput() these are start and end dates, by default set to “1900-01-01” and “2000-12-31” respectively.

In main_panel(), by calling the function plotOutput(), you define the type of output as a plot. You also define the plot’s outputId (as “series_bar_chart”) by which it can be identified in other parts of the app code.

The function series_histogram(), called by the server, replicates the R script for plots 4a and 4b (see the R script for Plots 13a and 13b, continued). Here it produces the chart depending on three parameters, series_subset (either “Professional” or “Personal,” the former being the default) as well as start_date and end_date with default values “1900-01-01” and “2000-01-01” respectively. After defining the lists “series_personal” and “series_professional,” set series_to_plot by default to “series_professional” (and, similarly, the plot title to “Professional papers”). If the argument series_subset is instead “Personal,” set series_to_plot to “series_personal” and the plot title to “Personal papers.” Next, produce the data table “counts_to_plot” with document counts for unique values in column “series_category” for the selected subset of series. Finally, plot the counts in a horizontal bar chart using the function ggplot().

In the second major component, the definition of function server(), you assign the plot produced by the function series_histogram() to output “series_bar_chart.”

Besides the argument data, assigned directly as the output of the function fread(), the remaining arguments are assigned from input, using the relevant input identifiers.

Finally, launch the R Shiny app by putting together the UI and server code with the function shinyApp().

Notes

1. Kevin Schindler and Will Grundy, Pluto and Lowell Observatory: A History of Discovery at Flagstaff (Charleston, SC: History Press, 2018); “Kevin Schindler,” Lowell Observatory, accessed July 10, 2020, https://lowell.edu/author/kevin-schindler/.
2. “Jackson-Gwilt Medal,” Royal Astronomical Society, accessed July 10, 2020, https://ras.ac.uk/awards-and-grants/awards/jackson-gwilt-medal.
3. “I Heart Pluto Festival,” Lowell Observatory, accessed July 10, 2020, https://lowell.edu/i-heart-pluto-festival.
4. “Clyde W. Tombaugh Collection,” NMSU Library, accessed July 12, 2020, https://lib.nmsu.edu/exhibits/tombaugh/bio.shtml.
5. New Mexico Museum of Space History home page, accessed July 12, 2020, https://www.nmspacemuseum.org/.
6. “Tombaugh Postdoctoral Fellowship,” NMSU Astronomy, accessed July 15, 2020, http://astronomy.nmsu.edu/fellowships-awards/tombaugh-fellowship/.
7. “Clyde W. Tombaugh,” New Mexico Museum of Space History, accessed July 16, 2020, https://www.nmspacemuseum.org/inductee/clyde-w-tombaugh/.
8. “Register of the Clyde W. Tombaugh Papers, 1908–2000,” Rocky Mountain Online Archive, accessed July 16, 2020, https://rmoa.unm.edu/docviewer.php?docId=nmlcu1ms0407.xml#hit.
9. Wikipedia, s.v. “Regular expression,” last modified July, 29, 2020, 06:09 (UTC), https://en.wikipedia.org/wiki/Regular_expression.
10. Google Maps Platform, accessed August 8, 2020, https://cloud.google.com/maps-platform/.
11. “Learn Shiny,” Shiny from RStudio, https://shiny.rstudio.com/tutorial/.
12. “Learn Shiny.”