As a NLP-centered software, gsitk has various functionalities for pre-processing tasks. Pre-processing is a central part of data munging in NLP, and more specifically, in sentiment analysis. In order to ease this kind of operation, gsitk offers the following type of pre-processers:

  • Simple: the simple and more efficient pre-processor. Indicated for processing large datasets, as it is the fastest. Based on regular expressions that parse English text.
  • Pre-process Twitter: a processor indicated for parsing Twitter text. Also based on regular expressions, extracts common emoji as special tokens, and transforms hashtags and mentions, normalizing them.
  • Normalize: An all-purpose pre-processor. It is not as efficient as the other options, and performs word tokenization based on NLTK.

Direct use

The more straight-forward way to use the pre-processing utilities is presented as follows:

from gsitk.preprocess import simple, pprocess_twitter, normalize

text = "My grandmother is an apple. Please, believe me!"
twitter_text = "@POTUS please let me enter to the USA #thanks"

print('simple', simple.preprocess(text))
print('twitter', pprocess_twitter.preprocess(twitter_text))
print('normalize', normalize.preprocess(text))

These lines of code would output the following. Note how the Twitter data is transformed, normalizing the mention to an user and a hashtag.

simple ['my', 'grandmother', 'is', 'an', 'apple', '.', 'please', ',', 'believe', 'me', '!']
twitter <user> please let me enter to the usa <allcaps> <hastag> thanks
normalize ['my', 'grandmother', 'is', 'an', 'apple', '.', 'please', ',', 'believe', 'me', '!']

Preprocessor interface

To facilitate the use of the preprocessing functions, gsitk offers an interface that is compatible with scikit-learn Pipelines: the Preprocessor. A simple script using this interface is:

from gsitk.preprocess import pprocess_twitter, Preprocessor

texts = [
    "@POTUS please let me enter to the USA #thanks",
    "If only Bradley's arm was longer. Best photo ever. #oscars"
# output:
array(['<user> please let me enter to the usa <allcaps> <hastag> thanks',
       "if only bradley's arm was longer. best photo ever. <hastag> oscars"],

The compatibility with scikit Pipelines can be used to ease the use of preprocessing. For example:

from sklearn.pipeline import Pipeline
from gsitk.preprocess import normalize, Preprocessor, JoinTransformer

texts = [
    "This cat is crazy, he is not on the mat!",
    "Will no one rid me of this turbulent priest?"

preprocessing_pipe = Pipeline([
    ('twitter', Preprocessor(normalize)),
    ('join', JoinTransformer())

# output:
['this cat is crazy , he is not on the mat !',
 'will no one rid me of this turbulent priest ?']

Stop word removal

Removing stop words is a pervasive task in NLP. gsitk includes a functionality for this, using the stop word collections in NLTK.

from gsitk.preprocess.stopwords import StopWordsRemover

texts = [
    "this cat is crazy , he is not on the mat !",
    "will no one rid me of this turbulent priest ?"

# output:
['cat crazy , mat !', 'one rid turbulent priest ?']

As it uses the NLTK stop word collections, several languages can be parsed, as in this Spanish example.

from gsitk.preprocess.stopwords import StopWordsRemover

texts = [
    "entre el clavel blanco y la rosa roja , su majestad escoja",
    "con diez cañones por banda viento en popa a toda vela",

# output:
['clavel blanco rosa roja , majestad escoja',
 'diez cañones banda viento popa toda vela']

Embeddings trick

The paper DepecheMood++: a Bilingual Emotion Lexicon Built Through Simple Yet Powerful Techniques (link here) introduces a technique called the Embeddings trick. In short, it consists on replacing certain words by others using a word embedding model. It is used to expand an existing emotion dictionary. In any way, we consider it is an useful technique, and has been implemented in gsitk, making it easier to replicate the mentioned paper.

This technique is used to avoid OOV (Out Of Vocabulary) issues when using a limited lexicon. Consider the following reduces sentiment analysis example:

from gsitk.preprocess.embeddings_trick import EmbeddingsTricker

et = EmbeddingsTricker(
    vocabulary=['my', 'cat', 'is', 'dog'],

ex_text = [
    ['my', 'cat', 'is', 'glad'],
    ['my', 'dog', 'is', 'saddening']

# output
[['my', 'cat', 'is', 'happy'], ['my', 'dog', 'is', 'sad']]

For more details on how to load and use the word embeddings for EmbeddingsTricker, see features.