smart-reading 1.1.4

An NLTK-based toolkit aimed at increasing the understanding of various texts.

Smart Reading

About

smart_reading is a Python module designed for increasing the understanding of various textforms by using natural language processing. It is heavily based on tools available from the Natural Language Toolkit (NLTK), which are used in various applications and provided with extensions.

Installation

The module is available for Python 2.7+, but recommended to run on Python 3+ for a more thorough unicode support and prettier graphs. Install via pip (or any other desired client):

$ pip install smart_reading

or by downloading the source code on PyPI or GitHub and running the following command in the root folder:

$ python setup.py install

Importing texts

The basic functionality of smart_reading is to import a given textfile into a smart_reading.book.Book object, which can be used to perform several developed analyses. Textfiles can be imported via the function smart_reading.book.load(filename). This function utilizes the functionality of the module textract to extract textual information of almost any given data structure, including .txt, .pdf, .epub and .docx. See its online documentation for more details on the inner workings of this module. When this module is not found on the system, the program continues with a limited functionality, in which only .txt files can be read. This limited functionality is added because it is experienced that the installation of textract does not always succeed. The user that is not able to install textract but does want to import other text formats is encouraged to build alternative pipelines to extract text into a .txt file, which in turn can be imported in the smart_reading.book.load function.

Additionally, a given string can be imported as an e-book via smart_reading.book.fromstring(text).

Three sample texts are included with different file structures, callable via the function smart_reading.book.sample:

>>> import smart_reading as sr
>>> sr.book.sample() # or sr.book.sample('txt')
Succesfully loaded 'Benn_Ch_II_The_Metaphysicians.txt' as an e-book
Total n.o. tokens: 10420
<smart_reading.book.Book instance at 0x105a546c8>
>>> sr.book.sample('pdf')
Succesfully loaded 'PhysRev.47.777.pdf' as an e-book
Total n.o. tokens: 3192
<smart_reading.book.Book instance at 0x110c0ebd8>
>>> sr.book.sample('epub') # this one takes a while to load
Succesfully loaded 'Galileo_The_Sidereal_Messenger.epub' as an e-book
Total n.o. tokens: 40372
<smart_reading.book.Book instance at 0x105a46e60>

Functionality

As mentioned, the given text are imported into a smart_reading.book.Book type object. The different tools that this object provides are listed below.

Concordance

A concordance is developed as an extension of the nltk.text.Text.concordance function that incorporates example 3.6 of the NLTK manual, such that it not only matches with exact copies of a given word, but also inflections:

>>> import smart_reading as sr
>>> bk = sr.book.sample()
Succesfully loaded 'Benn_Ch_II_The_Metaphysicians.txt' as an e-book
Total n.o. tokens: 10420
>>> bk.concordance('philosopher')
Displaying 17 of 17 matches:
nce of an independent income enabled the philosopher to live where he liked ; and
by our opinion of his metaphysics . As a philosopher Descartes has , to begin wit
r dazzle ; they could not convince . The philosophers professed to teach truth ; 
inctly are all true . In his other great philosophical work , the _Meditations_ ,
o his postulate of universal doubt , our philosopher argues from this to an imper
Here he agrees with another mathematical philosopher , Plato , who says the same 
at least one astronomer , who was also a philosopher , declared that the ultimate
 personality of God . SPINOZA . With the philosopher whom I have just named we co
sion of 500 florins on Spinoza , but the philosopher would accept no more than 30
l . To appreciate the work of the Hebrew philosopher , of the lonely muser , bred
 divine substance . In fact , the Hebrew philosopher does this , declaring boldly
peppers his pages . Yet , like the Greek philosophers , he is much more modern , 
 name of his great work that for him the philosophical problem is essentially a p
 . But he parts company with the English philosopher in his theory of what it mea
 alone , however , does not make a great philosopher ; character also is required
rity than any one utterance of any other philosopher ; but that fame is due to th
 work . On _à priori_ grounds the German philosopher seems to have an incontrover

In order to deal with inflections, the stemmer nltk.PorterStemmer is used by default. Other stemmers can be sent through the keyword stemmer when importing a textfile.

nltk.text.Text attribute

A smart_reading.book.Book object contains an attribute Text, which is an nltk.text.Text type object and as such includes all its attributes as developed by NLTK. These include finding collocations, similar words, and creating lexical disperion plots. See the NLTK API for its full documentation

>>> bk.Text.collocations()
fullest extent; infinite Power; material world; Princess Elizabeth;
external world; paramount object; supernatural revelation; two
attributes; necessarily exist; Queen Christina; early age; whole
universe; final causes; mathematical demonstration; metaphysical
system; perfection involves; best possible; many distinct;
mathematical method; divine substance
>>> bk.Text.dispersion_plot(['Descartes','Malebranche','Spinoza','Leibniz'])

The smart_reading.stats submodule

The smart_reading module comes with a stats submodule, which uses matplotlib.pyplot to create the several types of graps out of a given smart_reading.book.Book object, callable via the following functions:

smart_reading.stats.plot_noun_hist(book, no_nouns = 20, named_entities = True, exceptions = [], **kwargs)

Create a histogram of the most common nouns appearing in given text.

• no_nouns: Number of nouns that will be included in the graph (i.e. number of bars).
• named_entities: If false, this will exclude named entities like people and places that are recognized by the nltk.chunk.ne_chunk routine. This option is not supported in versions of Python 2.
• exceptions: An iterable of nouns that will be excluded from the analysis
• Further keyword arguments are passed to matplotlib.pyplot.fig.

>>> bk2 = sr.book.sample('pdf')
Succesfully loaded 'PhysRev.47.777.pdf' as an e-book
Total n.o. tokens: 3192
>>> sr.stats.plot_noun_hist(bk2)

smart_reading.stats.plot_network_graph(book, no_nouns = 10, treshold = 3, exclude_empty = True, named_entities = True, exceptions = [], **kwargs)

Create a network graph using the module networkx which depicts the relationship between frequently appearing nouns. The nouns appear as nodes, and edges are drawn between nouns if they appear frequently in the same sentences. A temperature color scheme is used on the edges to depict the frequency in which nouns appear together (red = very often, blue = a few times).

• no_nouns: Number of nouns that will be included in the graph (i.e. number of nodes). Can become less in the final result if exclude_empty is true, see below.
• treshold: The minimal number of sentences in which two given nouns have to appear in order for an edge to be drawn. Can be used to simplify graphs with a lot of edges.
• exclude_empty: if True, this will exclude the nodes from the graph that do not have edges. Note that this will reduce the number of nouns depicted, as declared in no_nouns above
• exceptions: an iterable of nouns that will be excluded from the analysis.
• Further keyword arguments are passed to matplotlib.pyplot.fig.

>>> sr.stats.plot_network_graph(bk2)

>>> bk3 = sr.book.sample('epub')
Succesfully loaded 'Galileo_The_Sidereal_Messenger.epub' as an e-book
Total n.o. tokens: 40372
>>> sr.stats.plot_network_graph(bk3, exceptions = ['Galileo'], treshold = 10)