WordsCounted

We are all in the gutter, but some of us are looking at the stars.

-- Oscar Wilde

WordsCounted is a Ruby NLP (natural language processor). WordsCounted lets you implement powerful tokensation strategies with a very flexible tokeniser class.

Are you using WordsCounted to do something interesting? Please tell me about it.

RubyDoc documentation.

Demo

Visit this website for one example of what you can do with WordsCounted.

Features

Out of the box, get the following data from any string or readable file, or URL:
- Token count and unique token count
- Token densities, frequencies, and lengths
- Char count and average chars per token
- The longest tokens and their lengths
- The most frequent tokens and their frequencies.
A flexible way to exclude tokens from the tokeniser. You can pass a string, regexp, symbol, lambda, or an array of any combination of those types for powerful tokenisation strategies.
Pass your own regexp rules to the tokeniser if you prefer. The default regexp filters special characters but keeps hyphens and apostrophes. It also plays nicely with diacritics (UTF and unicode characters): Bayrūt is treated as ["Bayrūt"] and not ["Bayr", "ū", "t"], for example.
Opens and reads files. Pass in a file path or a url instead of a string.

Installation

Add this line to your application's Gemfile:

gem 'words_counted'

And then execute:

$ bundle

Or install it yourself as:

$ gem install words_counted

Usage

Pass in a string or a file path, and an optional filter and/or regexp.

counter = WordsCounted.count(
  "We are all in the gutter, but some of us are looking at the stars."
)

# Using a file
counter = WordsCounted.from_file("path/or/url/to/my/file.txt")

.count and .from_file are convenience methods that take an input, tokenise it, and return an instance of WordsCounted::Counter initialized with the tokens. The WordsCounted::Tokeniser and WordsCounted::Counter classes can be used alone, however.

API

WordsCounted

WordsCounted.count(input, options = {})

Tokenises input and initializes a WordsCounted::Counter object with the resulting tokens.

counter = WordsCounted.count("Hello Beirut!")

Accepts two options: exclude and regexp. See Excluding tokens from the analyser and Passing in a custom regexp respectively.

WordsCounted.from_file(path, options = {})

Reads and tokenises a file, and initializes a WordsCounted::Counter object with the resulting tokens.

counter = WordsCounted.from_file("hello_beirut.txt")

Accepts the same options as .count.

Tokeniser

The tokeniser allows you to tokenise text in a variety of ways. You can pass in your own rules for tokenisation, and apply a powerful filter with any combination of rules as long as they can boil down into a lambda.

Out of the box the tokeniser includes only alpha chars. Hyphenated tokens and tokens with apostrophes are considered a single token.

#tokenise([pattern: TOKEN_REGEXP, exclude: nil])

tokeniser = WordsCounted::Tokeniser.new("Hello Beirut!").tokenise

# With `exclude`
tokeniser = WordsCounted::Tokeniser.new("Hello Beirut!").tokenise(exclude: "hello")

# With `pattern`
tokeniser = WordsCounted::Tokeniser.new("I <3 Beirut!").tokenise(pattern: /[a-z]/i)

See Excluding tokens from the analyser and Passing in a custom regexp for more information.

Counter

The WordsCounted::Counter class allows you to collect various statistics from an array of tokens.

#token_count

Returns the token count of a given string.

counter.token_count #=> 15

#token_frequency

Returns a sorted (unstable) two-dimensional array where each element is a token and its frequency. The array is sorted by frequency in descending order.

counter.token_frequency

[
  ["the", 2],
  ["are", 2],
  ["we",  1],
  # ...
  ["all", 1]
]

#most_frequent_tokens

Returns a hash where each key-value pair is a token and its frequency.

counter.most_frequent_tokens

{ "are" => 2, "the" => 2 }

#token_lengths

Returns a sorted (unstable) two-dimentional array where each element contains a token and its length. The array is sorted by length in descending order.

counter.token_lengths

[
  ["looking", 7],
  ["gutter",  6],
  ["stars",   5],
  # ...
  ["in",      2]
]

#longest_tokens

Returns a hash where each key-value pair is a token and its length.

counter.longest_tokens

{ "looking" => 7 }

#token_density([ precision: 2 ])

Returns a sorted (unstable) two-dimentional array where each element contains a token and its density as a float, rounded to a precision of two. The array is sorted by density in descending order. It accepts a precision argument, which must be a float.

counter.token_density

[
  ["are",     0.13],
  ["the",     0.13],
  ["but",     0.07 ],
  # ...
  ["we",      0.07 ]
]

#char_count

Returns the char count of tokens.

counter.char_count #=> 76

#average_chars_per_token([ precision: 2 ])

Returns the average char count per token rounded to two decimal places. Accepts a precision argument which defaults to two. Precision must be a float.

counter.average_chars_per_token #=> 4

#uniq_token_count

Returns the number of unique tokens.

counter.uniq_token_count #=> 13

Excluding tokens from the tokeniser

You can exclude anything you want from the input by passing the exclude option. The exclude option accepts a variety of filters and is extremely flexible.

A space-delimited string. The filter will normalise the string.
A regular expression.
A lambda.
A symbol that names a predicate method. For example :odd?.
An array of any combination of the above.

tokeniser =
  WordsCounted::Tokeniser.new(
    "Magnificent! That was magnificent, Trevor."
  )

# Using a string
tokeniser.tokenise(exclude: "was magnificent")
# => ["that", "trevor"]

# Using a regular expression
tokeniser.tokenise(exclude: /trevor/)
# => ["magnificent", "that", "was", "magnificent"]

# Using a lambda
tokeniser.tokenise(exclude: ->(t) { t.length < 4 })
# => ["magnificent", "that", "magnificent", "trevor"]

# Using symbol
tokeniser = WordsCounted::Tokeniser.new("Hello! محمد")
tokeniser.tokenise(exclude: :ascii_only?)
# => ["محمد"]

# Using an array
tokeniser = WordsCounted::Tokeniser.new(
  "Hello! اسماءنا هي محمد، كارولينا، سامي، وداني"
)
tokeniser.tokenise(
  exclude: [:ascii_only?, /محمد/, ->(t) { t.length > 6}, "و"]
)
# => ["هي", "سامي", "وداني"]

Passing in a custom regexp

The default regexp accounts for letters, hyphenated tokens, and apostrophes. This means twenty-one is treated as one token. So is Mohamad's.

/[\p{Alpha}\-']+/

You can pass your own criteria as a Ruby regular expression to split your string as desired.

For example, if you wanted to include numbers, you can override the regular expression:

counter = WordsCounted.count("Numbers 1, 2, and 3", pattern: /[\p{Alnum}\-']+/)
counter.tokens
#=> ["numbers", "1", "2", "and", "3"]

Opening and reading files

Use the from_file method to open files. from_file accepts the same options as .count. The file path can be a URL.

counter = WordsCounted.from_file("url/or/path/to/file.text")

Gotchas

A hyphen used in leu of an em or en dash will form part of the token. This affects the tokeniser algorithm.

counter = WordsCounted.count("How do you do?-you are well, I see.")
counter.token_frequency

[
  ["do",   2],
  ["how",  1],
  ["you",  1],
  ["-you", 1], # WTF, mate!
  ["are",  1],
  # ...
]

In this example -you and you are separate tokens. Also, the tokeniser does not include numbers by default. Remember that you can pass your own regular expression if the default behaviour does not fit your needs.

A note on case sensitivity

The program will normalise (downcase) all incoming strings for consistency and filters.

Roadmap

Ability to open URLs

def self.from_url
  # open url and send string here after removing html
end

Contributors

See contributors. Not listed there is [Dave Yarwood][1].

Contributing

Fork it
Create your feature branch (git checkout -b my-new-feature)
Commit your changes (git commit -am 'Add some feature')
Push to the branch (git push origin my-new-feature)
Create new Pull Request