Counting and mining research data with Unix
This lesson will look at how research data, when organised in a clear and predictable manner, can be counted and mined using the Unix shell. The lesson builds on the lessons “Preserving Your Research Data: Documenting and Structuring Data” and “Introduction to the Bash Command Line”. Depending on your confidence with the Unix shell, it can also be used as a standalone lesson or refresher.
Having accumulated research data for one project, a historian might ask different questions of that same data when returning to it during a subsequent project. If this data is spread across multiple files - a series of tabulated data, a set of transcribed text, a collection of images - it can be counted and mined using simple Unix commands.
The Unix shell gives you access to a range of powerful commands that can transform how you count and mine research data. This lesson will introduce you to a series of commands that use counting and mining of tabulated data, though they only scratch the surface of what the Unix shell can do. By learning just a few simple commands you will be able to undertake tasks that are impossible in Libre Office Calc, Microsoft Excel, or other similar spreadsheet programs. These commands can be easily extended for use with non-tabulated data.
This lesson will also demonstrate that the options for manipulating, counting and mining data available to you will often depend on the amount of metadata, or descriptive text, contained in the filenames of the data you are using as much as the range of Unix commands you have learnt to use. Thus, even if it is not a prerequisite of working with the Unix shell, taking the time to structure your research data and filenaming conventions in a consistent and predictable manner is certainly a significant step towards getting the most out of Unix commands and being able to count and mine your research data. For the value of taking the time to make your data consistent and predictable beyond matters of preservation, see “Preserving Your Research Data: Documenting and Structuring Data”.
Software and setup
OS X and Linux users will need to use their terminal shells to follow this lesson, as discussed in “Introduction to the Bash Command Line.”
This lesson was written using Git Bash 1.9.0 and the Windows 7 operating system. Equivalent file paths for OS X/Linux have been included where possible. Nonetheless, as commands and flags can change slightly between operating systems OS X/Linux users are referred to Deborah S. Ray and Eric J. Ray, “Unix and Linux: Visual Quickstart Guide”, 4th edition (2009) which covers interoperability in greater detail.
The files used in this lesson are available on “Figshare”. The data contains the metadata for journal articles categorised under ‘History’ in the British Library ESTAR database. The data is shared under a CC0 copyright waiver.
Download the required files, save them to your computer, and unzip them. If you do not have default software installed to interact with .zip files, we recommend 7-zip for this purpose. On Windows, we recommend unzipping the folder provided to your c: drive so the files are at
c:\proghist\. However, any location will work fine, but you may have to adjust your commands as you are following along with this lesson if you use a different location. On OS X or Linux, we similarly recommend unzipping them to your user directory, so that they appear at
/user/USERNAME/proghist/. In both cases, this means that when you open up a new terminal window, you can just type
cd proghist to move to the correct directory.
You will begin this lesson by counting the contents of files using the Unix shell. The Unix shell can be used to quickly generate counts from across files, something that is tricky to achieve using the graphical user interfaces (GUI) of standard office suites.
In Unix the
wc command is used to count the contents of a file or of a series of files.
Open the Unix shell and navigate to the directory that contains our data, the
data subdirectory of the
proghist directory. Remember, if at any time you are not sure where you are in your directory structure, type
pwd and use the
cd command to move to where you need to be. The directory structure here is slightly different between OS X/Linux and Windows: on the former, the directory is in a format such as
~/users/USERNAME/proghist/data and on Windows in a format such as
ls and then hit enter. This prints, or displays, a list that includes two files and a subdirectory.
The files in this directory are the dataset
2014-01_JA.csv that contains journal article metadata and a file containing documentation about
The subdirectory is named
derived_data. It contains four .tsv files derived from
2014-01_JA.csv. Each of these includes all data where a keyword such as
america appears in the ‘Title’ field of
derived_data directory also includes a subdirectory called
Note: CSV files are those in which the units of data (or cells) are separated by commas (comma-separated-values) and TSV files are those in which they are separated by tabs. Both can be read in simple text editors or in spreadsheet programs such as Libre Office Calc or Microsoft Excel.
Before you begin working with these files, you should move into the directory in which they are stored. Navigate to
c:\proghist\data\derived_data on Windows or
~/users/USERNAME/proghist/data/derived_data on OS X.
Now that you are here you can count the contents of the files.
The Unix command for counting is
wc -w 2014-01-31_JA_africa.tsv and hit enter. The flag
-w combined with
wc instructs the computer to print a word count, and the name of the file that has been counted, into the shell.
As was seen in “Introduction to the Bash Command Line”, flags such as
-w are an essential part of getting the most out of the Unix shell as they give you better control over commands.
If your research is more concerned with the number of entries (or lines) than the number of words, you can use the line count flag. Type
wc -l 2014-01-31_JA_africa.tsv and hit enter. Combined with
wc the flag
-l prints a line count and the name of the file that has been counted.
wc -c 2014-01-31_JA_africa.tsv and hit enter. This uses the flag
-c in combination with the command
wc to print a character count for
Note: OS X and Linux users should replace the
-c flag with
With these three flags, the most obvious thing historians can use
wc for is to quickly compare the shape of sources in digital format - for example word counts per page of a book, the distribution of characters per page across a collection of newspapers, the average line lengths used by poets. You can also use
wc with a combination of wildcards and flags to build more complex queries. Type
wc -l 2014-01-31_JA_a*.tsv and hit enter. This prints the line counts for
2014-01-31_JA_america.tsv, offering a simple means of comparing these two sets of research data. Of course, it may be faster to compare the line count for the two documents in Libre Office Calc, Microsoft Excel, or a similar spreadsheet program. But when wishing to compare the line count for tens, hundreds, or thousands of documents, the Unix shell has a clear speed advantage.
Moreover, as our datasets increase in size you can use the Unix shell to do more than copy these line counts by hand, by the use of print screen, or by copy and paste methods. Using the
> redirect operator you can export your query results to a new file. Type
wc -l 2014-01-31_JA_a*.tsv > results/2014-01-31_JA_a_wc.txt and hit enter. This runs the same query as before, but rather than print the results within the Unix shell it saves the results as
2014-01-31_JA_a_wc.txt. By prefacing this with
results/ it moves the .txt file to the
results sub-directory. To check this, navigate to the
results subdirectory, hit enter, type
ls, and hit enter again to see this file listed within
c:\proghist\data\derived_data\results on Windows or
/users/USERNAME/proghist/data/derived_data/results on OS X/Linux.
The Unix shell can do much more than count the words, characters, and lines within a file. The
grep command (meaning ‘global regular expression print’) is used to search across multiple files for specific strings of characters. It is able to do so much faster than the graphical search interface offered by most operating systems or office suites. And combined with the
> operator, the
grep command becomes a powerful research tool can be used to mine your data for characteristics or word clusters that appear across multiple files and then export that data to a new file. The only limitations here are your imagination, the shape of your data, and - when working with thousands or millions of files - the processing power at your disposal.
To begin using
grep, first navigate to the
derived_data directory (
cd ..). Here type
grep 1999 *.tsv and hit enter. This query looks across all files in the directory that fit the given criteria (the .tsv files) for instances of the string, or character cluster, ‘1999’. It then prints them within the shell.
Note: there is a large amount of data to print, so if you get bored hit
ctrl+c to cancel the action. Ctrl+c is used to cancel any process in the Unix shell.
Press the up arrow once in order to cycle back to your most recent action. Amend
grep 1999 *.tsv to
grep -c 1999 *.tsv and hit enter. The shell now prints the number of times the string 1999 appeared in each .tsv file. Cycle to the previous line again and amend this to
grep -c 1999 2014-01-31_JA_*.tsv > results/2014-01-31_JA_1999.txt and hit enter. This query looks for instances of the string ‘1999’ across all documents that fit the criteria and saves them as
2014-01-31_JA_1999.txt in the
Strings need not be numbers.
grep -c revolution 2014-01-31_JA_america.tsv 2014-02-02_JA_britain.tsv, for example, counts the instances of the string
revolution within the defined files and prints those counts to the shell. Run this and then amend it to
grep -ci revolution 2014-01-31_JA_america.tsv 2014-02-02_JA_britain.tsv. This repeats the query, but prints a case insensitive count (including instances of both
Revolution). Note how the count has increased nearly 30 fold for those journal article titles that contain the keyword ‘revolution’. As before, cycling back and adding
> results/, followed by a filename (ideally in .txt format), will save the results to a data file.
You can also use
grep to create subsets of tabulated data. Type
grep -i revolution 2014-01-31_JA_america.tsv 2014-02-02_JA_britain.tsv > YEAR-MONTH-DAY_JA_america_britain_i_revolution.tsv (where
YEAR-MONTH-DAY is the date you are completing this lesson) and hit enter. This command looks in both of the defined files and exports any lines containing
revolution (without regard to case) to the specified .tsv file.
The data has not been saved to to the
results directory because it isn’t strictly a result; it is derived data. Depending on your research project it may be easier to save this to another subdirectory. For now have a look at this file to verify its contents and when you are happy, delete it using the
rm command. Note: the
rm common is very powerful and should be used with caution. Please refer to “Introduction to the Bash Command Line” for instructions on how to use this command correctly.
Finally, you can use another flag,
-v, to exclude data elements when using the
grep command. Type
grep -iv revolution 2014*_JA_a*.tsv > 2014_JA_iv_revolution.csv and hit enter. This query looks in the defined files (three in total) and exports all lines that do not contain
Note that you have transformed the data from one format to another - from .tsv to .csv. Often there is a loss of data structure when undertaking such transformations. To observe this for yourself, run
grep -iv revolution 2014*_JA_a*.tsv > 2014_JA_iv_revolution.tsv and open both the .csv and .tsv files in Libre Office Calc, Microsoft Excel, or a similar spreadsheet program. Note the differences in column delineation between the two files.
Within the Unix shell you can now:
- use the
wccommand with the flags
-lto count the words and lines in a file or a series of files.
- use the redirector and structure
> subdirectory/filenameto save results into a subdirectory.
- use the
grepcommand to search for instances of a string.
- use with
-cflag to count instances of a string, the
-iflag to return a case insensitive search for a string, and the
-vflag to exclude a string from the results.
- combine these commands and flags to build complex queries in a way that suggests the potential for using the Unix shell to count and mine your research data and research projects.
In this lesson you have learnt to undertake some basic file counting, to query across research data for common strings, and to save results and derived data. Though this lesson is restricted to using the Unix shell to count and mine tabulated data, the processes can be easily extended to free text. For this we recommend two guides written by William Turkel:
- William Turkel, ‘Basic Text Analysis with Command Line Tools in Linux’ (15 June 2013)
- William Turkel, ‘Pattern Matching and Permuted Term Indexing with Command Line Tools in Linux’ (20 June 2013)
As these recommendations suggest, the present lesson only scratches the surface of what the Unix shell environment is capable of. It is hoped, however, that this lesson has provided a taster sufficient to prompt further investigation and productive play.
For many historians, the full potential of these tools may only emerge upon embedding these skills into a real research project. Once your research grows, and, with it, your research data, being able to manipulate, count and mine thousands of files will be extremely useful. For if you choose to build on this lesson and investigate the Unix shell further you will find that even a large collection of files which do not contain any alpha-numeric data elements, such as image files, can be easily sorted, selected and queried in the Unix shell.
James Baker and Ian Milligan , "Counting and mining research data with Unix," Programming Historian (20 September 2014), http://programminghistorian.org/lessons/research-data-with-unix