Skip to main content

A fast framework for pre-processing (Cleaning text, Reduction of vocabulary, Feature extraction and Vectorization). Implemented with parallel processing using custom number of processes.

Project description

Preprocess NLP Text

Framework Description

A simple and fast framework for

  • Preprocessing or Cleaning of text
  • Extracting top words or reduction of vocabulary
  • Feature Extraction
  • Word Vectorization

Uses parallel execution by leveraging the multiprocessing library in Python for cleaning of text, extracting top words and feature extraction modules. Contains both sequential and parallel ways (For less CPU intensive processes) for preprocessing text with an option of user-defined number of processes.

PS: There is no multi-processing support for word vectorization

  • Cleaning Text - Clean text with various defined stages implemented using standardized techniques in Natural Language Processing (NLP)
  • Vocab Reduction - Find the top words in the corpus, lets you choose a threshold to consider the words that can stay in the corpus and replaces the others
  • Feature Extraction - Extract features from corpus of text using SpaCy
  • Word Vectorization - Simple code to convert words to vectors (TFIDF, Word2Vec, GloVe) using Scikit-learn and Gensim

Preprocess/Cleaning Module

Uses nltk for few of the stages defined below. Various stages of cleaning include:

Stage Description
remove_tags_nonascii Remove HTML tags, emails, URLs, non-ascii characters and converts accented characters
lower_case Converts the text to lower_case
expand_contractions Expands the word contractions
remove_punctuation Remove punctuation from text, but sentences are seperated by ' . '
remove_esacape_chars Remove escapse characters like \n, \t etc
remove_stopwords Remove stopwords using nltk python
remove_numbers Remove all digits in the text
lemmatize Uses WordNetLemmatizer to lemmatize text
stemming Uses SnowballStemmer for stemming of text
min_word_len Minimum word length to keep in text

Reduction of Vocabulary

Shortlists top words based on the percentage as input. Replaces the words not shortlisted and replaces them efficienctly. Also, supports parallel and sequential processing.

Feature Extraction Module

Uses Spacy Pipe module to avoid unnecessary parsing to increase speed. Various stages of feature extraction include:

Stage Description
nouns Extract the list of Nouns from the given string
verbs Extract the list of Verbs from the given string
adjs Extract the list of Adjectives from the given string
noun_phrases Extract the list of Noun Phrases (Noun chunks) from the given string
keywords Uses YAKE for extracting keywords from text
ner Extracts Person, Location and Organization as named entities
numbers Extracts all digits in the text

Word Vectorization

Functions written in python to convert words to vectors using libraries like Scikit-Learn and Gensim. Contains four vectorization techniques like CountVectorizer (Bag of Words Model), TFIDF-Vectorizer, Word2Vec and GloVe. Also contains others features to get the top words according to IDF Scores, similar words with similarity scores and average sentence-wise vectors.


Code - Components

Various Python files and their purposes are mentioned here:


How to run

  1. pip install -r requirements.txt
  2. Import preprocess_nlp.py and use the functions preprocess_nlp(for sequential) and asyn_call_preprocess(for parallel) as defined in notebook
  3. Import vocab_elimination_nlp.py and use functions as defined in the notebook Vocab_Elimination_Example_Notebook.ipynb
  4. Import feature_extraction.py and use functions as defined in notebook Feature_Extraction_Example_Notebook.ipynb
  5. Import vectorization_nlp.py and use functions as defined in notebook Vectorization_Example_Notebook.ipynb

Sequential & Parallel Processing

  1. Sequential - Processes records in a sequential order, does not consume a lot of CPU Memory but is slower compared to Parallel processing
  2. Parallel - Can create multiple processes (customizable/user-defined) to preprocess text parallelly, Memory intensive and faster

Refer the code for Docstrings and other function related documentation.
Cheers :)

Project details


Release history Release notifications | RSS feed

This version

0.1

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

preprocess_nlp-0.1-py3-none-any.whl (3.7 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page