bagofwords 1.0.4

The main goal this Python module is to provide functions to apply Text Classification.

Introduction

A Python module that allows you to create and manage a collection of occurrence counts of words without regard to grammar. The main purpose is provide a set of classes to manage several document classifieds by category in order to apply Text Classification.

You can make use via API or via Command Line. For example, you can generate your classified documents (learn) via Command Line and after via API classify an input document.

Third parties modules

Module uses three third parties modules

• stop_words

• pystemmer

• six

The first module is used in stop_words filter, the second module is used in stemming filter. If you don’t use these two filters, you don’t need install them.

Installation

Install it via pip

$ [sudo] pip install bagofwords

Or download zip and then install it by running

$ [sudo] python setup.py install

You can test it by running

$ [sudo] python setup.py test

Uninstallation

$ [sudo] pip uninstall bagofwords

Python API

Methods

• document_classifier(document, **classifieds) Text classification based on an implementation of Naive Bayes

Module contains two main classes DocumentClass and Document and four secondary classes BagOfWords, WordFilters, TextFilters and Tokenizer

Main classes

• DocumentClass Implementing a bag of words collection where all the bags of words are the same category, as well as a bag of words with the entire collection of words. Each bag of words has an identifier otherwise it’s assigned an calculated identifier. Retrieves the text of a file, folder, url or zip, and also allows save or retrieve the collection in json format.

• Document Implementing a bag of words where all words are of the same category. Retrieves the text of a file, folder, url or zip, and also allows save or retrieve the Document in json format.

Secondary classes

• BagOfWords Implementing a bag of words with their frequency of usages.

• TextFilters Filters for transforming a text. It’s used in Tokenizer class. Including filters upper lower invalid_chars and html_to_text

• WordFilters Filters for transforming a set of words. It’s used in Tokenizer class. Including filters stemming stopwords and normalize

• Tokenizer Allows to break a string into tokens (set of words). Optionally allows you to set filters before (TextFilters) and after (WordFilters) breaking the string into tokens.

Subclasses

• Tokenizer subclasses DefaultTokenizer SimpleTokenizer and HtmlTokenizer that implements the more common filters and overwriting after_tokenizer and berofe_tokenizer methods

• Document subclasses DefaultDocument SimpleDocument and HtmlDocument

• DocumentClass subclasses DefaultDocumentClass SimpleDocumentClass and HtmlDocumentClass

Command Line Tool

usage: bow [-h] [--version] {create,learn,show,classify} ...

Manage several document to apply text classification.

positional arguments:
  {create,learn,show,classify}
    create              create classifier
    learn               add words learned a classifier
    show                show classifier info
    classify            Naive Bayes text classification

optional arguments:
  -h, --help            show this help message and exit
  --version             show version and exit

Create Command

usage: bow create [-h] [--lang-filter LANG_FILTER]
                  [--stemming-filter STEMMING_FILTER]
                  {text,html} filename

positional arguments:
  {text,html}           filter type
  filename              file to be created where words learned are saved

optional arguments:
  -h, --help            show this help message and exit
  --lang-filter LANG_FILTER
                        language text where remove empty words
  --stemming-filter STEMMING_FILTER
                        number loops of lemmatizing

Learn Command

usage: bow learn [-h] [--file FILE [FILE ...]] [--dir DIR [DIR ...]]
                 [--url URL [URL ...]] [--zip ZIP [ZIP ...]] [--no-learn]
                 [--rewrite] [--list-top-words LIST_TOP_WORDS]
                 filename

positional arguments:
  filename              file to write words learned

optional arguments:
  -h, --help            show this help message and exit
  --file FILE [FILE ...]
                        filenames to learn
  --dir DIR [DIR ...]   directories to learn
  --url URL [URL ...]   url resources to learn
  --zip ZIP [ZIP ...]   zip filenames to learn
  --no-learn            not write to file the words learned
  --rewrite             overwrite the file
  --list-top-words LIST_TOP_WORDS
                        maximum number of words to list, 50 by default, -1
                        list all

Show Command

usage: bow show [-h] [--list-top-words LIST_TOP_WORDS] filename

positional arguments:
  filename              filename

optional arguments:
  -h, --help            show this help message and exit
  --list-top-words LIST_TOP_WORDS
                        maximum number of words to list, 50 by default, -1
                        list all

Classify Command

usage: bow classify [-h] [--file FILE] [--url URL] [--text TEXT]
                    classifiers [classifiers ...]

positional arguments:
  classifiers  classifiers

optional arguments:
  -h, --help   show this help message and exit
  --file FILE  file to classify
  --url URL    url resource to classify
  --text TEXT  text to classify

Example

Previously you need to download a spam corpus enron-spam dataset. For example you can download a compressed file that includes a directory with 1500 spam emails and a directory with 4012 ham emails.

http://www.aueb.gr/users/ion/data/enron-spam/preprocessed/enron3.tar.gz

Now we will create the spam and ham classifiers

$ bow create text spam
* filename: spam
* filter:
    type: DefaultDocument
    lang: english
    stemming: 1
* total words: 0
* total docs: 0

$ bow create text ham
* filename: ham
* filter:
    type: DefaultDocument
    lang: english
    stemming: 1
* total words: 0
* total docs: 0

It’s time to learn

$ bow learn spam --dir enron3/spam

current
=======
* filename: spam
* filter:
    type: DefaultDocument
    lang: english
    stemming: 1
* total words: 0
* total docs: 0

updated
=======
* filename: spam
* filter:
    type: DefaultDocument
    lang: english
    stemming: 1
* total words: 223145
* total docs: 1500
* pos | word (top 50)                       | occurrence |       rate
  --- | ----------------------------------- | ---------- | ----------
    1 | "                                   |       2438 | 0.01092563
    2 | subject                             |       1662 | 0.00744807
    3 | compani                             |       1659 | 0.00743463
    4 | s                                   |       1499 | 0.00671761
    5 | will                                |       1194 | 0.00535078
    6 | com                                 |        978 | 0.00438280
    7 | statement                           |        935 | 0.00419010
    8 | secur                               |        908 | 0.00406910
    9 | inform                              |        880 | 0.00394362
   10 | e                                   |        802 | 0.00359408
   11 | can                                 |        798 | 0.00357615
   12 | http                                |        779 | 0.00349100
   13 | pleas                               |        743 | 0.00332967
   14 | invest                              |        740 | 0.00331623
   15 | de                                  |        739 | 0.00331175
   16 | o                                   |        733 | 0.00328486
   17 | 1                                   |        732 | 0.00328038
   18 | 2                                   |        709 | 0.00317731
   19 | stock                               |        700 | 0.00313697
   20 | price                               |        664 | 0.00297564
  ....

$ bow learn ham --dir enron3/ham

current
=======
* filename: ham
* filter:
    type: DefaultDocument
    lang: english
    stemming: 1
* total words: 0
* total docs: 0

updated
=======
* filename: ham
* filter:
    type: DefaultDocument
    lang: english
    stemming: 1
* total words: 1293023
* total docs: 4012
* pos | word (top 50)                       | occurrence |       rate
  --- | ----------------------------------- | ---------- | ----------
    1 | enron                               |      29805 | 0.02305063
    2 | s                                   |      22438 | 0.01735313
    3 | "                                   |      15712 | 0.01215137
    4 | compani                             |      12039 | 0.00931074
    5 | said                                |       9470 | 0.00732392
    6 | will                                |       8862 | 0.00685371
    7 | 2001                                |       8293 | 0.00641365
    8 | subject                             |       7167 | 0.00554282
    9 | 1                                   |       5887 | 0.00455290
   10 | trade                               |       5718 | 0.00442220
   11 | energi                              |       5599 | 0.00433016
   12 | market                              |       5498 | 0.00425205
   13 | new                                 |       5278 | 0.00408191
   14 | 2                                   |       4742 | 0.00366737
   15 | dynegi                              |       4651 | 0.00359700
   16 | stock                               |       4594 | 0.00355291
   17 | 10                                  |       4545 | 0.00351502
   18 | year                                |       4517 | 0.00349336
   19 | power                               |       4503 | 0.00348254
   20 | share                               |       4393 | 0.00339746
 ....

Finally, we can classify a text file or url

$ bow classify spam ham --text "company"

* classifier                          |       rate
  ----------------------------------- | ----------
  ham                                 | 0.87888743
  spam                                | 0.12111257

$ bow classify spam ham --text "new lottery"

* classifier                          |       rate
  ----------------------------------- | ----------
  spam                                | 0.96633627
  ham                                 | 0.03366373

$ bow classify spam ham --text "Subject: a friendly professional online pharmacy focused on you !"

* classifier                          |       rate
  ----------------------------------- | ----------
  spam                                | 0.99671480
  ham                                 | 0.00328520

You should know that it is also possible to classify from python code

import bow

spam = bow.Document.load('spam')
ham = bow.Document.load('ham')
dc = bow.DefaultDocument()

dc.read_text("company")
result = bow.document_classifier(dc, spam=spam, ham=ham)

print result

Result

[('ham', 0.8788874288217258), ('spam', 0.12111257117827418)]

Others examples

Join several bag of words

from bow import BagOfWords

a = BagOfWords('car', 'chair', 'chicken')
b = BagOfWords({'chicken':2}, ['eye', 'ugly'])
c = BagOfWords('plane')

print a + b + c
print a - b - c

Result

{'eye': 1, 'car': 1, 'ugly': 1, 'plane': 1, 'chair': 1, 'chicken': 3}
{'car': 1, 'chair': 1}

HTML document class

from bow import HtmlDocumentClass

html_one = '''
<!DOCTYPE html>
<html lang="en">
<head>
  <title>bag of words demo</title>
  <link rel="stylesheet" href="css/mycss.css">
  <script src="js/myjs.js"></script>
</head>
<body>
  <style> #demo {background: #c00; color: #fff; padding: 10px;}</style>
  <!--my comment section -->
  <h2>This is a demo</h2>
  <p id="demo">This a text example of my bag of words demo!</p>
  I hope this demo is useful for you
  <script type="text/javascript"> alert('But wait, it\'s a demo...');</script>
</body>
</html>
'''

html_two = '''
<!DOCTYPE html>
<html lang="en">
<head> </head>
<body> Another silly example. </body>
</html>
'''

dclass = HtmlDocumentClass(lang='english', stemming=0)
dclass(id_='doc1', text=html_one)
dclass(id_='doc2', text=html_two)
print 'docs \n', dclass.docs
print 'total \n', dclass
print 'rates \n', dclass.rates

Result

>>>
docs
{
 'doc2': {u'silly': 1, u'example': 1, u'another': 1},
 'doc1': {u'useful': 1, u'text': 1, u'bag': 2, u'words': 2, u'demo': 4, u'example': 1, u'hope': 1}
}
total
{
 u'useful': 1, u'another': 1, u'text': 1, u'bag': 2, u'silly': 1, u'words': 2,
 u'demo': 4, u'example': 2, u'hope': 1
}
rates
{
 u'useful': 0.06666666666666667, u'another': 0.06666666666666667, u'text': 0.06666666666666667,
 u'bag': 0.13333333333333333, u'silly': 0.06666666666666667, u'words': 0.13333333333333333,
 u'demo': 0.26666666666666666, u'example': 0.13333333333333333, u'hope': 0.06666666666666667
}
>>>

License

MIT License, see LICENSE