数据, 术→技巧

Python解析拼音输入法词库

钱魏Way · · 1,745 次浏览

目前类似搜狗输入法、百度输入法等最大的亮点是有较好的词库,而这些词库除了用在搜索上还可以用在及其场景,比如说分词。目前每种输入法的词库都有自己的格式,今天我们就来研究下,如何反编译这些词库,将这些词库使用到其他的场景。

搜狗输入法词库解析

搜狗输入法提供的词库下载下来是scel格式的,在使用前把他转成txt,以下为转换方法:

# -*- coding: utf-8 -*-

import struct
import sys

# 搜狗的scel词库就是保存的文本的unicode编码,每两个字节一个字符(中文汉字或者英文字母)
#找出其每部分的偏移位置即可
#主要两部分
#1.全局拼音表,貌似是所有的拼音组合,字典序
#       格式为(index,len,pinyin)的列表
#       index: 两个字节的整数 代表这个拼音的索引
#       len: 两个字节的整数 拼音的字节长度
#       pinyin: 当前的拼音,每个字符两个字节,总长len
#
#2.汉语词组表
#       格式为(same,py_table_len,py_table,{word_len,word,ext_len,ext})的一个列表
#       same: 两个字节 整数 同音词数量
#       py_table_len:  两个字节 整数
#       py_table: 整数列表,每个整数两个字节,每个整数代表一个拼音的索引
#
#       word_len:两个字节 整数 代表中文词组字节数长度
#       word: 中文词组,每个中文汉字两个字节,总长度word_len
#       ext_len: 两个字节 整数 代表扩展信息的长度,好像都是10
#       ext: 扩展信息 前两个字节是一个整数(不知道是不是词频) 后八个字节全是0
#
#      {word_len,word,ext_len,ext} 一共重复same次 同音词 相同拼音表

#拼音表偏移,
startPy = 0x1540

#汉语词组表偏移
startChinese = 0x2628

#全局拼音表
GPy_Table = {}

#解析结果
#元组(词频,拼音,中文词组)的列表
GTable = []


def byte2str(data):
    '''''将原始字节码转为字符串'''
    i = 0
    length = len(data)
    ret = u''
    while i < length:
        x = data[i] + data[i + 1]
        t = unichr(struct.unpack('H', x)[0])
        if t == u'\r':
            ret += u'\n'
        elif t != u' ':
            ret += t
        i += 2
    return ret


#获取拼音表
def getPyTable(data):
    if data[0:4] != "\x9D\x01\x00\x00":
        return None
    data = data[4:]
    pos = 0
    length = len(data)
    while pos < length:
        index = struct.unpack('H', data[pos] + data[pos + 1])[0]
        #print index,
        pos += 2
        l = struct.unpack('H', data[pos] + data[pos + 1])[0]
        #print l,
        pos += 2
        py = byte2str(data[pos:pos + l])
        #print py
        GPy_Table[index] = py
        pos += l


#获取一个词组的拼音
def getWordPy(data):
    pos = 0
    length = len(data)
    ret = u''
    while pos < length:
        index = struct.unpack('H', data[pos] + data[pos + 1])[0]
        ret += GPy_Table[index]
        pos += 2
    return ret


#获取一个词组
def getWord(data):
    pos = 0
    length = len(data)
    ret = u''
    while pos < length:
        index = struct.unpack('H', data[pos] + data[pos + 1])[0]
        ret += GPy_Table[index]
        pos += 2
    return ret


#读取中文表
def getChinese(data):
    #import pdb
    #pdb.set_trace()

    pos = 0
    length = len(data)
    while pos < length:
        #同音词数量
        same = struct.unpack('H', data[pos] + data[pos + 1])[0]
        #print '[same]:',same,

        #拼音索引表长度
        pos += 2
        py_table_len = struct.unpack('H', data[pos] + data[pos + 1])[0]
        #拼音索引表
        pos += 2
        py = getWordPy(data[pos: pos + py_table_len])

        #中文词组
        pos += py_table_len
        for i in xrange(same):
            #中文词组长度
            c_len = struct.unpack('H', data[pos] + data[pos + 1])[0]
            #中文词组
            pos += 2
            word = byte2str(data[pos: pos + c_len])
            #扩展数据长度
            pos += c_len
            ext_len = struct.unpack('H', data[pos] + data[pos + 1])[0]
            #词频
            pos += 2
            count = struct.unpack('H', data[pos] + data[pos + 1])[0]

            #保存
            GTable.append((count, py, word))

            #到下个词的偏移位置
            pos += ext_len


def deal(file_name):
    print '-' * 60
    f = open(file_name, 'rb')
    data = f.read()
    f.close()

    if data[0:12] != "\x40\x15\x00\x00\x44\x43\x53\x01\x01\x00\x00\x00":
        print "确认你选择的是搜狗(.scel)词库?"
        sys.exit(0)
    #pdb.set_trace()

    print "词库名:", byte2str(data[0x130:0x338])  #.encode('GB18030')
    print "词库类型:", byte2str(data[0x338:0x540])  #.encode('GB18030')
    print "描述信息:", byte2str(data[0x540:0xd40])  #.encode('GB18030')
    print "词库示例:", byte2str(data[0xd40:startPy])  #.encode('GB18030')

    getPyTable(data[startPy:startChinese])
    getChinese(data[startChinese:])


if __name__ == '__main__':

    #将要转换的词库添加在这里就可以了
    f = "搜狗标准词库.scel".decode("utf-8")

    deal(f)

    #保存结果
    f = open('sougou.txt', 'w')
    for count, py, word in GTable:
        #GTable保存着结果,是一个列表,每个元素是一个元组(词频,拼音,中文词组),有需要的话可以保存成自己需要个格式
        #我没排序,所以结果是按照上面输入文件的顺序
        f.write(unicode('{%(count)s}' % {'count': count} + py + ' ' + word).encode('GB18030'))  #最终保存文件的编码,可以自给改
        f.write('\n')
    f.close()

其他参考资料:

百度输入法词库解析

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import struct
from tools import *


class bdict(BaseDictFile):
    def __init__(self):
        BaseDictFile.__init__(self)
        # 文件头
        self.head = 'biptbdsw'
        # 文件终点偏移
        self.offset = 0x60
        self.end_position = 0
        # 词表偏移
        self.dict_start = 0x350
        self.shengmu = [ "c", "d", "b", "f", "g", "h", "ch", "j", "k", "l", "m", "n", "", "p", "q", "r", "s", "t", "sh", "zh", "w", "x", "y", "z"]
        self.yunmu = [ "uang", "iang", "iong", "ang", "eng", "ian", "iao", "ing", "ong", "uai", "uan", "ai", "an", "ao", "ei", "en", "er", "ua", "ie", "in", "iu", "ou", "ia", "ue", "ui", "un", "uo", "a", "e", "i", "o", "u", "v"]

    def _get_word_len(self, data, pos = 0):
        # 单词长度
        length = struct.unpack('I', data[pos:pos+4])[0]
        char = struct.unpack('B', data[pos+4])[0]
        pure_english = False
        if 0x41 <= char <= 0x7a:
            pure_english = True
        #print(repr(data[pos:pos+length*3+1]))
        return (length, pure_english)

    def _get_word(self, data, length = 0, pure_english = False):
        pos = 0
        word = Word()
        pinyin = []
        for i in xrange(length):
            char = struct.unpack('B', data[pos])[0]
            pos += 1
            if pure_english:
                # 如果读取到的首个hex值落在字母区域(0x41 ~ 0x7a)内,说明是纯英文,后面的内容直接读取即可
                pinyin.append(chr(char).lower())
                word.value += chr(char)
            else:
                if char == 0xff:
                    # 声母部分如果是'\xff',说明这是中英文混输的英文字母,不需要做拼音词表转换,直接读取韵母部分即可
                    sm = ''
                    ym = struct.unpack('c', data[pos])[0]
                    pos += 1
                    pinyin.append('' + ym)
                else:
                    sm = char
                    ym = struct.unpack('B', data[pos])[0]
                    pos += 1
                    pinyin.append(self.shengmu[sm] + self.yunmu[ym])
#                    try:
#                        pinyin.append(self.shengmu[sm] + self.yunmu[ym])
#                    except IndexError:
#                        print(repr(data))
#                        raise IndexError
        if pure_english:
            word.pinyin = ' '.join(pinyin)
        else:
            word.pinyin = ' '.join(pinyin)
            hanzi = byte2str(data[pos:pos+ length*2])
            pos = pos+ length*2
            word.value = hanzi.encode('utf-8')
        return word

    def get_dict_info(self, data):
        self.end_position = struct.unpack('I', data[self.offset:self.offset+4])[0]


    def load(self, filename):
        f = open(filename,'rb')
        data = f.read()
        f.close()

        if data[0:8] != self.head:
            print "It's not a bdict file"
            sys.exit(1)

        return self.read(data)

    def read(self, data):
        self.get_dict_info(data)

        pos = self.dict_start
        while pos < self.end_position:
#            print(pos, self.end_position)
#            if pos >= self.end_position:
#                break
            # 这里的长度是算字数的,实际长度是 拼音数(length * 2) + 字符长度(length * 2)
            length, pure_english = self._get_word_len(data, pos = pos)
            pos += 4
            #print('*'*60)
            #print(repr(data[pos:pos+length*4]))
            if pure_english:
                word = self._get_word(data[pos:pos+length], length = length, pure_english = True)
                pos = pos + length
            else:
                word = self._get_word(data[pos:pos+length*4], length = length)
                pos = pos + length * 4
            if word.value:
                if self.dictionary.has_key(word.pinyin):
                    self.dictionary[word.pinyin].append(word)
                else:
                    self.dictionary[word.pinyin] = []
                    self.dictionary[word.pinyin].append(word)

        return self.dictionary

其他参考资料:

QQ拼音输入法

JAVA版本:

/*  Copyright (c) 2010 Xiaoyun Zhu
 * 
 *  Permission is hereby granted, free of charge, to any person obtaining a copy  
 *  of this software and associated documentation files (the "Software"), to deal  
 *  in the Software without restriction, including without limitation the rights  
 *  to use, copy, modify, merge, publish, distribute, sublicense, and/or sell  
 *  copies of the Software, and to permit persons to whom the Software is  
 *  furnished to do so, subject to the following conditions:
 *  
 *  The above copyright notice and this permission notice shall be included in  
 *  all copies or substantial portions of the Software.
 *  
 *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR  
 *  IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,  
 *  FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE  
 *  AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER  
 *  LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,  
 *  OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN  
 *  THE SOFTWARE.  
 */

import java.io.ByteArrayOutputStream;
import java.io.FileOutputStream;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;
import java.nio.channels.Channels;
import java.nio.channels.FileChannel;
import java.util.Arrays;
import java.util.zip.InflaterOutputStream;

/**
 * QQ Pinyin IME QPYD File Reader
 * 
 * <pre>
 * QPYD Format overview:
 * 
 * General Information:
 * - Chinese characters are all encoded with UTF-16LE.
 * - Pinyin are encoded in ascii (or UTF-8).
 * - Numbers are using little endian byte order.
 * 
 * QPYD hex analysis:
 * - 0x00 QPYD file identifier
 * - 0x38 offset of compressed data (word-pinyin-dictionary)
 * - 0x44 total words in qpyd
 * - 0x60 start of header information
 * 
 * Compressed data analysis:
 * - zip/standard (beginning with 0x789C) is used in (all analyzed) qpyd files
 * - data is divided in two parts
 * -- 1. offset and length information (16 bytes for each pinyin-word pair)
 *       0x06 offset points to first pinyin
 *       0x00 length of pinyin
 *       0x01 length of word
 * -- 2. actual data
 *       Dictionary data has the form ((pinyin)(word))* with no separators.
 *       Data can only be read using offset and length information.
 * 
 * </pre>
 * 
 */
public class QQPinyinQpydReader {
  public static void main(final String[] args) throws IOException {
    // download from http://dict.py.qq.com/list.php
    final String qqydFile = "D:\\test.qpyd";

    // read qpyd into byte array
    final ByteArrayOutputStream dataOut = new ByteArrayOutputStream();

    try (RandomAccessFile file = new RandomAccessFile(qqydFile, "r"); final FileChannel fChannel = file.getChannel();) {
      fChannel.transferTo(0, fChannel.size(), Channels.newChannel(dataOut));
    }

    // qpyd as bytes
    final ByteBuffer dataRawBytes = ByteBuffer.wrap(dataOut.toByteArray());
    dataRawBytes.order(ByteOrder.LITTLE_ENDIAN);

    System.out.println("文件: " + qqydFile);

    // read info of compressed data
    final int startZippedDictAddr = dataRawBytes.getInt(0x38);
    final int zippedDictLength = dataRawBytes.limit() - startZippedDictAddr;

    // qpys as UTF-16LE string
    final String dataString = new String(Arrays.copyOfRange(dataRawBytes.array(), 0x60, startZippedDictAddr), "UTF-16LE");

    // print header
    System.out.println("名称:" + QQPinyinQpydReader.substringBetween(dataString, "Name: ", "\r\n"));
    System.out.println("类型:" + QQPinyinQpydReader.substringBetween(dataString, "Type: ", "\r\n"));
    System.out.println("子类型:" + QQPinyinQpydReader.substringBetween(dataString, "FirstType: ", "\r\n"));
    System.out.println("词库说明:" + QQPinyinQpydReader.substringBetween(dataString, "Intro: ", "\r\n"));
    System.out.println("词库样例:" + QQPinyinQpydReader.substringBetween(dataString, "Example: ", "\r\n"));
    System.out.println("词条数:" + dataRawBytes.getInt(0x44));

    // read zipped qqyd dictionary into byte array
    dataOut.reset();
    try (InflaterOutputStream inflater = new InflaterOutputStream(dataOut);) {
      Channels.newChannel(inflater).write(ByteBuffer.wrap(dataRawBytes.array(), startZippedDictAddr, zippedDictLength));
    }

    // uncompressed qqyd dictionary as bytes
    final ByteBuffer dataUnzippedBytes = ByteBuffer.wrap(dataOut.toByteArray());
    dataUnzippedBytes.order(ByteOrder.LITTLE_ENDIAN);

    // for debugging: save unzipped data to *.unzipped file
    try (FileOutputStream out = new FileOutputStream(qqydFile + ".unzipped");) {
      Channels.newChannel(out).write(dataUnzippedBytes);
      System.out.println("压缩数据:0x" + Integer.toHexString(startZippedDictAddr) + " (解压前:" + zippedDictLength + " B, 解压后:" + dataUnzippedBytes.limit() + " B)");
    }

    // stores the start address of actual dictionary data
    int unzippedDictStartAddr = -1;
    final byte[] byteArray = dataUnzippedBytes.array();
    dataUnzippedBytes.position(0);
    while ((unzippedDictStartAddr == -1) || (dataUnzippedBytes.position() < unzippedDictStartAddr)) {
      // read word
      final int pinyinLength = dataUnzippedBytes.get() & 0xff;
      final int wordLength = dataUnzippedBytes.get() & 0xff;
      dataUnzippedBytes.getInt(); // garbage
      final int pinyinStartAddr = dataUnzippedBytes.getInt();
      final int wordStartAddr = pinyinStartAddr + pinyinLength;

      if (unzippedDictStartAddr == -1) {
        unzippedDictStartAddr = pinyinStartAddr;
        System.out.println("词库地址(解压后):0x" + Integer.toHexString(unzippedDictStartAddr) + "\n");
      }

      final String pinyin = new String(Arrays.copyOfRange(byteArray, pinyinStartAddr, pinyinStartAddr + pinyinLength), "UTF-8");
      final String word = new String(Arrays.copyOfRange(byteArray, wordStartAddr, wordStartAddr + wordLength), "UTF-16LE");
      System.out.println(word + "\t" + pinyin);
    }
  }

  public static final String substringBetween(final String text, final String start, final String end) {
    final int nStart = text.indexOf(start);
    final int nEnd = text.indexOf(end, nStart + 1);
    if ((nStart != -1) && (nEnd != -1)) {
      return text.substring(nStart + start.length(), nEnd);
    } else {
      return null;
    }
  }
}

其他参考资料:

常见输入法词库下载地址

发表回复

您的电子邮箱地址不会被公开。 必填项已用*标注