xinci 新词 & 抽词

xinci is a Python interface for chinese words extraction & new words extraction. [https://pypi.org/project/xinci/]

c++ version see fast-xinci

Requirements

Python >= 2.7

Installation

1. using pip

pip install xinci

2. using setup.py

git clone https://github.com/Lapis-Hong/xinci.git 
cd xinci 
pip setup.py install

Usage

This package has two main use cases: words extraction and find new words.

1. command line

cd xinci
python word_extraction.py 

or

./run.sh

2. python package

import xinci

# if you want to see logging events.
import logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s : %(levelname)s : %(message)s')

# init default dictionary or user dic,
dic = xinci.Dictionary()
# load vocab, vocab is a python set.
vocab = dic.load()  # or dic.dictionary
print(vocab)

# add words to dic
dic.add(['神马'])  # or dic.add_from_file('user.dic')
# remove words from dic
dic.remove(['神马'])  # or dic.remove_from_file('user.dic')

# extract new words, xc is a set
xc = xinci.extract('corpus.txt')
for w in xc:
    print(w)
# extract all words, c is a set
c = xinci.extract('corpus.txt', all_words=True)
for w in xc:
    print(w)

result

发现5个新词如下:
@新词	@词频
祛斑	13
后再	7
今日头条	9
洗净切	7
蛋液	9

Notes: Iteratively add "not seems to new words" in result to common dic will improve a lot.

API documentation

xc = xinci.extract(params)

List of available params and their default value:

corpus_file:           string, input corpus file (required)
common_words_file:     string, common words dic file [common.dic]
min_candidate_len:     int, min candidate word length [2]
max_candidate_len:     int, max candidate word length [5]
least_cnt_threshold:   int, least word count to extract [5]
solid_rate_threshold:  float, solid rate threshold [0.018]
entropy_threshold:     float, entropy threshold [1.92]
all_words: bool,       set True to extract all words mode [False]
save_file: string,     output file [None]

References

The code is based on this java version [https://github.com/GeorgeBourne/grid]