Skip to content

Natural Language Processing

Natural Language Processing (NLP) is the subfield of machine learning that works with human language data. The nlp module contains four classes that help to convert raw text to meaningful numeric values, ready to be ingested by a model. ATOM uses the nltk library for the majority of its NLP processes.

The text documents are expected to be provided in a column of the dataframe named corpus (the name is case-insensitive). Only the corpus is changed by the transformers, leaving the rest of the columns as is. This mechanism allows atom to combine datasets containing a text corpus with other non-text features. If an array is provided as input, it should consist of only one feature containing the text (one document per row). ATOM will then automatically convert the array to a dataframe with the desired column name. Documents are expected to be strings or sequences of words. Click here for an example using text data.

Note

All of atom's NLP methods automatically adopt the relevant transformer attributes (verbose, logger) from atom. A different choice can be added as parameter to the method call, e.g. atom.tokenize(verbose=0).

Info

ATOM doesn't do topic modelling! The module's goal is to help process text documents into features that can be used for supervised learning.


Text cleaning

Text data is rarely clean. Whether it's scraped from a website or inferred from paper documents, it's always populated with irrelevant information for the model, such as email addresses, HTML tags, numbers or punctuation marks. Use the TextCleaner class to clean the corpus from such noise. It can be accessed from atom through the textclean method. Use the class' parameters to choose which transformations to perform. The available steps are:

  • Decode unicode characters to their ascii representations.
  • Convert all characters to lower case.
  • Drop email addresses from the text.
  • Drop URL links from the text.
  • Drop HTML tags from the text.
  • Drop emojis from the text.
  • Drop numbers from the text.
  • Drop punctuations from the text.


Tokenization

Some text processing algorithms, like stemming or lemmatization, require the corpus to be made out of tokens, instead of strings, in order to know what to consider as words. Tokenization is used to achieve this. It separates every document into a sequence of smaller units. In this case, the words.

Sometimes, words have a different meaning on their own than when combined with adjacent words. For example, the word new has a completely different meaning when the word york is directly after it than when it's not. These combinations of two words are called bigrams. When there are three words, they are called trigrams, and with four words quadgrams.

The Tokenizer class converts a document into a sequence of words, and can create the most frequent bigrams, trigrams and quadgrams. It can be accessed from atom through the tokenize method.


Text Normalization

Normalization for texts is a process that converts a list of words to a more uniform standard. This is useful to reduce the amount of different information that the computer has to deal with, and therefore improves efficiency. The goal of normalization techniques like stemming and lemmatization is to reduce inflectional and related forms of a word to a common base form.

Normalize the words in the corpus using the TextNormalizer class. It can be accessed from atom through the textnormalize method.


Vectorization

Text data cannot be fed directly to the algorithms themselves, as most of them expect numerical feature vectors with a fixed size, rather than words in the text documents with variable length. Vectorization is the general process of turning a collection of text documents into numerical feature vectors. You can apply it to the corpus using the Vectorizer class. It can be accessed from atom through the vectorize method.

Info

All strategies can utilize GPU speed-up. Click here for further information about GPU acceleration.


Bag of Words
The Bag of Words (BOW) strategy applies tokenization, counting and normalization to the corpus. Documents are described by word occurrences while completely ignoring the relative position information of the words in the document. The created columns are named with the words they are embedding with the prefix corpus_. Read more in sklearn's documentation.


TF-IDF
In a large text corpus, some words will be very present (e.g. “the”, “a”, “is” in English), hence carrying very little meaningful information about the actual contents of the document. If we were to feed the direct count data directly to a classifier, those very frequent terms would shadow the frequencies of rarer, yet more interesting, terms. Use the TF-IDF strategy to re-weight the count features into floating point values. The created columns are named with the words they are embedding with the prefix corpus_. Read more in sklearn's documentation.


Hashing
The larger the corpus, the larger the vocabulary will grow and thus increasing the number of features and memory use. Use the Hashing strategy to hash the words to a specified number of features. The created features are named hash0, hash1, etc... Read more in sklearn's documentation.