天天看点

[python]将搜狗(sogou)的细胞词库转换为mmseg的词库

From: https://github.com/aboutstudy/scel2mmseg

------------------------------------------------------------

将搜狗(sogou)的细胞词库转换为mmseg的词库

功能:

  • scel2mmseg.py: 将.scel文件转换为mmseg格式的.txt文件

    使用方法: python scel2mmseg.py a.scel a.txt

    批量转换方法:python scel2mmseg.py scel文件目录 a.txt

    说明:新增加的所有词的词频都为1,对于格式的解释如下:[摘自 http://jjw.in/server/226 ]

    每条记录分两行。其中,第一行为词项,其格式为:[词条]\t[词频率]。需要注意的是,对于单个字后面跟这个字作单字成词的频率,这个频率需要在大量的预先切分好的语料库中进行统计,用户增加或删除词时,一般不需要修改这个数值;对于非单字词,词频率处必须为1。第二行为占位项,是由于LibMMSeg库的代码是从Coreseek其他的分词算法库(N-gram模型)中改造而来的,在原来的应用中,第二行为该词在各种词性下的分布频率。LibMMSeg的用户只需要简单的在第二行处填”x:1″即可
  • mergedict.py: 将mmseg的多个.txt文件合并为一个.txt

    使用方法: python mergedict.py unigram.txt b.txt c.txt new.txt

    说明: .txt可以使mmseg格式的,也可以是每行一个词的格式(这样词频默认为1)

    注意:因为merge的时候会判重,一个词在前面出现过,就不会追加到新产生的文件中,所以要将unigram.txt放到最前面

------------------------------------------------------------

scel2mmseg.py:

------------------------------------------------------------

import struct
import os, sys, glob

def read_utf16_str (f, offset=-1, len=2):
    if offset >= 0:
        f.seek(offset)
    str = f.read(len)
    return str.decode('UTF-16LE')

def read_uint16 (f):
    return struct.unpack ('<H', f.read(2))[0]

def get_word_from_sogou_cell_dict (fname):
    f = open (fname, 'rb')
    file_size = os.path.getsize (fname)
    
    hz_offset = 0
    mask = struct.unpack ('B', f.read(128)[4])[0]
    if mask == 0x44:
        hz_offset = 0x2628
    elif mask == 0x45:
        hz_offset = 0x26c4
    else:
        sys.exit(1)
    
    title   = read_utf16_str (f, 0x130, 0x338  - 0x130)
    type    = read_utf16_str (f, 0x338, 0x540  - 0x338)
    desc    = read_utf16_str (f, 0x540, 0xd40  - 0x540)
    samples = read_utf16_str (f, 0xd40, 0x1540 - 0xd40)
    
    py_map = {}
    f.seek(0x1540+4)
    
    while 1:
        py_code = read_uint16 (f)
        py_len  = read_uint16 (f)
        py_str  = read_utf16_str (f, -1, py_len)
    
        if py_code not in py_map:
            py_map[py_code] = py_str
    
        if py_str == 'zuo':
            break
    
    f.seek(hz_offset)
    while f.tell() != file_size:
        word_count   = read_uint16 (f)
        pinyin_count = read_uint16 (f) / 2
    
        py_set = []
        for i in range(pinyin_count):
            py_id = read_uint16(f)
            py_set.append(py_map[py_id])
        py_str = "'".join (py_set)

        for i in range(word_count):
            word_len = read_uint16(f)
            word_str = read_utf16_str (f, -1, word_len)
            f.read(12) 
            yield py_str, word_str

    f.close()

def showtxt (records):
    for (pystr, utf8str) in records:
        print len(utf8str), utf8str

def store(records, f):
    for (pystr, utf8str) in records:
        f.write("%s\t1\n" %(utf8str.encode("utf8")))
        f.write("x:1\n")

def main ():
	if len (sys.argv) != 3:
		print "Unknown Option \n usage: python %s file.scel new.txt" %(sys.argv[0])
		exit (1)
	
	#Specify the param of scel path as a directory, you can place many scel file in this dirctory, the this process will combine the result in one txt file
	if os.path.isdir(sys.argv[1]):
		for fileName in glob.glob(sys.argv[1] + '*.scel'):
			print fileName
			generator = get_word_from_sogou_cell_dict(fileName)
			with open(sys.argv[2], "a") as f:
				store(generator, f)

	else:
		generator = get_word_from_sogou_cell_dict (sys.argv[1])
		with open(sys.argv[2], "w") as f:
			store(generator, f)
			#showtxt(generator)

if __name__ == "__main__":
    main()
           

------------------------------------------------------------

继续阅读