A Multilingual Latent Dirichlet Allocation (LDA) Pipeline with Stop Words Removal, n-gram features, and Inverse Stemming, in Python.

Last update: Oct 07, 2022

Overview

Multilingual Latent Dirichlet Allocation (LDA) Pipeline

This project is for text clustering using the Latent Dirichlet Allocation (LDA) algorithm. It can be adapted to many languages provided that the Snowball stemmer, a dependency of this project, supports it.

Usage

from artifici_lda.lda_service import train_lda_pipeline_default


FR_STOPWORDS = [
    "le", "les", "la", "un", "de", "en",
    "a", "b", "c", "s",
    "est", "sur", "tres", "donc", "sont",
    # even slang/texto stop words:
    "ya", "pis", "yer"]
# Note: this list of stop words is poor and is just as an example.

fr_comments = [
    "Un super-chat marche sur le trottoir",
    "Les super-chats aiment ronronner",
    "Les chats sont ronrons",
    "Un super-chien aboie",
    "Deux super-chiens",
    "Combien de chiens sont en train d'aboyer?"
]

transformed_comments, top_comments, _1_grams, _2_grams = train_lda_pipeline_default(
    fr_comments,
    n_topics=2,
    stopwords=FR_STOPWORDS,
    language='french')

print(transformed_comments)
print(top_comments)
print(_1_grams)
print(_2_grams)

Output:

array([[0.14218195, 0.85781805],
       [0.11032992, 0.88967008],
       [0.16960695, 0.83039305],
       [0.88967041, 0.11032959],
       [0.8578187 , 0.1421813 ],
       [0.83039303, 0.16960697]])

['Un super-chien aboie', 'Les super-chats aiment ronronner']

[[('chiens', 3.4911404011996545), ('super', 2.5000203653313933)],
 [('chats',  3.4911393765493255), ('super', 2.499979634668601 )]]

[[('super chiens', 2.4921035508342464)],
 [('super chats',  2.492102155345991 )]]

How it works

See Multilingual-LDA-Pipeline-Tutorial for an exhaustive example (intended to be read from top to bottom, not skimmed through). For more explanations on the Inverse Lemmatization, see Stemming-words-from-multiple-languages.

Supported Languages

Those languages are supported:

Danish
Dutch
English
Finnish
French
German
Hungarian
Italian
Norwegian
Porter
Portuguese
Romanian
Russian
Spanish
Swedish
Turkish

You need to bring your own list of stop words. That could be achieved by computing the Term Frequencies on your corpus (or on a bigger corpus of the same language) and to use some of the most common words as stop words.

Dependencies and their license

numpy==1.14.3           # BSD-3-Clause and BSD-2-Clause BSD-like and Zlib
scikit-learn==0.19.1    # BSD-3-Clause
PyStemmer==1.3.0        # BSD-3-Clause and MIT
snowballstemmer==1.2.1  # BSD-3-Clause and BSD-2-Clause
translitcodec==0.4.0    # MIT License
scipy==1.1.0            # BSD-3-Clause and MIT-like

Unit tests

Run pytest with ./run_tests.sh. Coverage:

----------- coverage: platform linux, python 3.6.7-final-0 -----------
Name                                       Stmts   Miss  Cover
--------------------------------------------------------------
artifici_lda/__init__.py                       0      0   100%
artifici_lda/data_utils.py                    39      0   100%
artifici_lda/lda_service.py                   31      0   100%
artifici_lda/logic/__init__.py                 0      0   100%
artifici_lda/logic/count_vectorizer.py         9      0   100%
artifici_lda/logic/lda.py                     23      7    70%
artifici_lda/logic/letter_splitter.py         36      4    89%
artifici_lda/logic/stemmer.py                 60      3    95%
artifici_lda/logic/stop_words_remover.py      61      5    92%
--------------------------------------------------------------
TOTAL                                        259     19    93%

License

This project is published under the MIT License (MIT).

Coded by Guillaume Chevalier at Neuraxio Inc.

A Multilingual Latent Dirichlet Allocation (LDA) Pipeline with Stop Words Removal, n-gram features, and Inverse Stemming, in Python.

Related tags

Overview

Multilingual Latent Dirichlet Allocation (LDA) Pipeline

Usage

How it works

Supported Languages

Dependencies and their license

Unit tests

License

Owner

Artifici Online Services inc.

MiCECo - Misskey Custom Emoji Counter

Negative sampling for solving the unlabeled entity problem in NER. ICLR-2021 paper: Empirical Analysis of Unlabeled Entity Problem in Named Entity Recognition.

Japanese NLP Library

A modular Karton Framework service that unpacks common packers like UPX and others using the Qiling Framework.

Lumped-element impedance calculator and frequency-domain plotter.

PyTorch code for EMNLP 2019 paper "LXMERT: Learning Cross-Modality Encoder Representations from Transformers".

AutoGluon: AutoML for Text, Image, and Tabular Data

PatrickStar enables Larger, Faster, Greener Pretrained Models for NLP. Democratize AI for everyone.

DLO8012: Natural Language Processing & CSL804: Computational Lab - II

SAINT PyTorch implementation

Include MelGAN, HifiGAN and Multiband-HifiGAN, maybe NHV in the future.

中文无监督SimCSE Pytorch实现

TweebankNLP - Pre-trained Tweet NLP Pipeline (NER, tokenization, lemmatization, POS tagging, dependency parsing) + Models + Tweebank-NER

Idea is to build a model which will take keywords as inputs and generate sentences as outputs.

Mastering Transformers, published by Packt

Fake news detector filters - Smart filter project allow to classify the quality of information and web pages

NLP - Machine learning

Transformer Based Korean Sentence Spacing Corrector

Train GPT-3 model on V100(16GB Mem) Using improved Transformer.