A simple Thai wordcut written in Python, based on Maximum Matching algorithm by S. Manabu . Uses Lexitron (by NECTEC) dictionary as default
Please note: This project is under development and should not be use in production , all function and interface are subject to change. If you have issue or suggestion please feel free to ask, contribution is also very welcome :)
pip install pythaiwordcut
or
git clone https://github.com/zenyai/pythaiwordcut.git
python setup.py install
import pythaiwordcut as pwt
pt = pwt.wordcut(removeRepeat=True, stopDictionary="<full path to txt file>", removeSpaces=True, minLength=1, stopNumber=False, removeNonCharacter=False, caseSensitive=True, ngram=(1, 2), negation=False)
print "|".join(pt.segment(u'ทดสอบการตัดคำ'))
- removeRepeat: remove intention insertion spelling error such as (สบายยยยยย)
- stopDictionary: remove word that exist in this specify text file (one word one line)
- removeSpaces: remove blank space
- minLength: minimum length of each word
- stopNumber: remove number if exist
- removeNonCharacter: remove character that are not Thai or English character
- caseSensitive: if set to false, will remove stop word without regarding the case
- ngram: Add word ngram from (1, 2)
- negation: If set to true, then it will add NOT_ to every word after negation word and space