Big Data, Big Corpus, and Bi-Grams: Calculating Literary Complexity

```Big Data, Big Corpus,
and Bigrams:
Calculating Literary
Complexity
Nathaniel Husted [email protected]
You too can be a Big Data
Scientist!
Terminology: Big Data
Not a new concept (never believe marketers)
Moving target
Data sets large enough to cause extra considerations for processing and
storage
Terminology: Big Corpora
Corpus (Plural: Corpora) – A sample set of texts for natural language
processing.
Big Corpus – A very large, gigabyte level, set of texts.
◦ Example: Corpus of Contemporary American English
Terminology: Bigrams
The Quick Brown Fox Leaves.
Also known as a Digram or n-gram for n=2.
Terminology: Bigrams
The Quick Brown Fox Leaves.
Also known as a Digram or n-gram for n=2.
Terminology: Bigrams
The Quick Brown Fox Leaves.
Also known as a Digram or n-gram for n=2.
Terminology: Bigrams
The Quick Brown Fox Leaves.
Also known as a Digram or n-gram for n=2.
Terminology: Literary
Complexity
The Complexity of a Story.
◦
◦
◦
◦
◦
Qualitative
How intertwined are the plot lines
How deep are the themes
How rich are the characters
How much attention it takes on the part of the read to comprehend the
whole
Examples of Complex Litearture:
◦ Finnegan’s Wake by James Joyce
◦ Foucault’s Pendulum by Umberto Eco
Terminology: A Little Graph
Theory
Vertex
Undirected Edge
Directed Edge
Loop
Let’s Put Them All Together…
Structural Complexity
How can we quantitatively measure the complexity of a novel?
◦ Structural Complexity!
◦ Biologists use structure to measure the complexity of molecules
◦ System Scientists use it to measure the complexity of networks
What is Structural Complexity?
◦ The amount of information contained in the relationship between elements
of a network.
Metrics of Structural
Complexity
Normalized Edge Complexity (NEC)
◦ How many unique bigrams there were versus the theoretical maximum.
Average Edge Complexity (AEC)
◦ Average number of unique bigrams per word.
Shannon Information (SI)
Vertex degree magnitude-based Information (IVD)
http://www.vcu.edu/csbc/pdfs/quantitative_measures.pdf
Structural Complexity In
Literature:
Bigrams as Structural Cues
To use our structural complexity measures, we must “graph” our novel.
Bigrams provide a clear notion of a “graph edge”
Structural Complexity In
Literature:
Bigrams as Structural Cues
The Quick Brown Fox Leaves The House.
The
Quick
House
Leaves
Brown
Fox
How do we implement all
these concepts?
Python!
◦NetworkX
◦NLTK
◦XMLTree
SQLite
(xargs)
What is our process?
1.
Choose our Corpus
2.
Organize our Corpus
3.
Parse our Corpus
4.
Analyze our Graphs
5.
Process our Results
Choosing our Corpus
Project Gutenberg to the Rescue
◦
◦
◦
◦
Tens of thousands of texts
Most, if not all, are in text formats (ASCII, ISO, UTF-8)
Public Domain!
Number of works: 19852
Number of authors: 7049
https://www.cs.Indiana.edu/~nhusted/project_source/pgdvd-encorpus.tar.bz2
Organizing our Corpus
Project Gutenberg provides a RDF Card Catalogue of their library.
Querying a 250+ MB RDF file with RDF libraries is SLOW.
Parsing with Python’s xml.etree.cElementTree is fast!
Due to Unicode Characters, Python 3 is a must.
Storing results in SQLite give us a compact, quickly searchable, format.
Parsing our Corpus in to
Graphs!
Python, NetworkX, and NLTK to the rescue.
NLTK allows quick parsing of the novels.
NetworkX provides the easy to use graph library with algorithms.
Analyzing Our Graphs’
Structural Complexity
IVD
AEV
Storing and Analyzing the
Results
Store the results in SQLite
◦ Conveniently searchable, still.
Use R for Statistical Analysis
◦ Personal Preference
So what can we say about
Structural Complexity?
It seems to have dropped in
the late 1800s
Structural Complexity is
Analogous to Literary Complexity
Determine authors who have literature deemed “complex”
Publisher’s Weekly Top 10 Most Difficult Books:
http://www.publishersweekly.com/pw/by-topic/industry-news/tipsheet/article/53409-the-top-10-most-difficult-books.html
Structural Complexity is
Analogous to Literary Complexity
http://www.publishersweekly.com/pw/by-topic/industrynews/tip-sheet/article/53409-the-top-10-most-difficultbooks.html
Structural Complexity is
Analogous to Literary Complexity
http://www.publishersweekly.com/pw/by-topic/industrynews/tip-sheet/article/53409-the-top-10-most-difficultbooks.html
Structural Complexity is
Analogous to Literary Complexity
http://www.publishersweekly.com/pw/by-topic/industrynews/tip-sheet/article/53409-the-top-10-most-difficultbooks.html
Conclusions
Structural Complexity is analogous to qualitative measurements of
literary complexity
Structural Complexity even allows comparison of novels to other
structures such as DNA and protein-protein sequences
Results are preliminary
◦ Data is not Gaussian
◦ Still some catalog creation errors
◦ “Big Data” is still sparse
Big Conclusion: Open Source
Science!
Results are Creative Commons!
Code is GPL V3!
Dataset is public domain!
You can do your own analysis!
http://cgi.cs.indiana.edu/~nhusted/dokuwiki/doku.php?id=projects:gra
phalyzer
https://github.iu.edu/nhusted/GutenbergGraphalyzer
You too can be a Big Data
Scientist!
```