tkseem

tkseem (تقسيم) is a tokenization library that encapsulates different approaches for tokenization and preprocessing of Arabic text.

Documentation

Please visit readthedocs for the full documentation.

Installation

pip install tkseem

Usage

Tokenization

import tkseem as tk
tokenizer = tk.WordTokenizer()
tokenizer.train('samples/data.txt')

tokenizer.tokenize("السلام عليكم")
tokenizer.encode("السلام عليكم")
tokenizer.decode([536, 829])

Caching

tokenizer.tokenize(open('data/raw/train.txt').read(), use_cache = True)

Save and Load

import tkseem as tk

tokenizer = tk.WordTokenizer()
tokenizer.train('samples/data.txt')

# save the model
tokenizer.save_model('vocab.pl')

# load the model
tokenizer = tk.WordTokenizer()
tokenizer.load_model('vocab.pl')

Model Agnostic

import tkseem as tk
import time 
import seaborn as sns
import pandas as pd

def calc_time(fun):
    start_time = time.time()
    fun().train()
    return time.time() - start_time

running_times = {}

running_times['Word'] = calc_time(tk.WordTokenizer)
running_times['SP'] = calc_time(tk.SentencePieceTokenizer)
running_times['Random'] = calc_time(tk.RandomTokenizer)
running_times['Disjoint'] = calc_time(tk.DisjointLetterTokenizer)
running_times['Char'] = calc_time(tk.CharacterTokenizer)

Notebooks

We show how to use tkseem to train some nlp models.

Name	Description	Notebook
Demo	Explain the syntax of all tokenizers.
Sentiment Classification	WordTokenizer for processing sentences and then train a classifier for sentiment classification.
Meter Classification	CharacterTokenizer for meter classification using bidirectional GRUs.
Translation	Seq-to-seq model with attention.

Citation

@misc{tkseem2020,
  author = {Zaid Alyafeai and Maged Saeed},
  title = {tkseem: A Tokenization Library for Arabic.},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/ARBML/tkseem}}
}

Contribution

This is an open source project where we encourage contributions from the community.

License

MIT license.