crf-pos 2.2.2

Persian Part-of-Speech tagger framework

Persian Parts-of-Speech tagger

This repository contains Persian Part of Speech tagger based on Conditional Random Fields and a native Text Normalizer.

Table of Contents

1. TO-DO
2. Installation
   1. Using Pip
   2. From Source
   3. On CoLab
3. Usage
4. Implementation Details
5. Evaluation
6. How To Contribute

TO-DO:

• CRF tagger commit#64
• Wapiti tagger commit#56
• Native Normalizer pull#4
• UnitTesting commit#127
• CI/CD pull#5
• Scrutinize Coverage issue#8
• Documentation pull#9
• Improve Coverage pull#9
• Smooth Installation issue#12 pull#13
• Excel code quality pull#11
• Adding documentation and flowchart of the code.
• CircleCI CI/CD Pipeline Config issue#14

Installation:

Using Pip

! pip install crf_pos

From Source

$ git clone https://github.com/MohammadForouhesh/crf-pos-persian 
$ cd crf-pos-persian
$ python setup.py install

On CoLab

! pip install git+https://github.com/MohammadForouhesh/crf-pos-persian.git

Usage

from crf_pos.pos_tagger.wapiti import WapitiPosTagger
pos_tagger = WapitiPosTagger()
tokens = 'او رئیس‌جمهور حجتالاسلاموالمسلمین ابرهیم رئیسی رئیس جمهور ایران اسلامی می باشد'
pos_tagger[tokens]

[1]: 
[('او', 'PRO'),
('رئیس\u200cجمهور', 'N'),
('حجت\u200cالاسلام\u200cوالمسلمین', 'N'),
('ابرهیم', 'N'),
('رئیسی', 'N'),
('رئیس\u200cجمهور', 'N'),
('ایران', 'N'),
('اسلامی', 'ADJ'),
('می\u200cباشد', 'V')]

Implementation Details

Evaluation

Test and training is perfomed on Mojgan Seraji's Uppsala Persian Corpus

Part-of-Speech	Description	precision	recall	f1-score	support
N	Noun	0.985	0.970	0.977	186585
P	Preposition	0.998	0.998	0.998	89450
V	Verb	0.999	0.999	0.999	87762
ADV	Adverb	0.976	0.972	0.974	15983
FW	Foreign Word	0.989	0.992	0.991	2784
DET	Determiner	0.973	0.977	0.975	19786
ADJ	Adjective	0.978	0.975	0.977	61526
INT	Interjection	1.000	1.000	1.000	73
CONJ	Conjunction	0.996	0.997	0.997	74796
PRO	Pronoun	0.973	0.974	0.973	23094
NUM	Numeral	0.988	0.992	0.990	24864
avg/total	-	0.985	0.985	0.985	586703

How To Contribute

1. Report any encountered error trough [BUG]
2. Report if Normalizer mis-out half-space correction trough [ZWNJ]