系統：win7 32位

分詞軟體：PyNLPIR

整合開發環境（IDE）：Pycharm

功能：實現多級文字預處理全過程，包括文字分詞，過濾停用詞，詞頻統計，特徵選擇，文字表示，並將結果匯出為WEKA能夠處理的.arff格式。

直接上程式碼：

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

"""
功能：PyNLPIR文字預處理
過程：文字分詞，過濾停用詞，詞頻統計，特徵選擇，文字表示
時間：2016年8月25日10:52:43
"""

import pynlpir
import codecs
import math

pynlpir.open()

#文字分詞
typelist = [u"財經",u"IT",u"健康",u"體育",u"旅遊",u"教育",u"招聘",u"文化",u"軍事"]

typetxt = codecs.open('C:\\Users\\Administrator\\Desktop\\txttype.txt', 'a', encoding='utf-8')
wordseg_result = codecs.open('C:\\Users\\Administrator\\Desktop\\wordseg_result.txt', 'a',encoding='utf-8')

allresult = []
for j in range(1,10):
    for i in range(10,510):
        typetxt.write(typelist[j-1] + "\n")

        s = ""
        singletext_result = []
        print  (u'正在對第 %s 個資料夾的第 %s 個文字進行分詞處理.....'%(j,i))
        f = codecs.open('C:\\Users\\Administrator\\Desktop\\textmining_experiment2\\Word Segment\\traintxt500\\%d\\%d.txt' % (j,i),'r',"gb18030")
        for line in f:
            s += line.strip().encode('utf-8')

        for item in pynlpir.segment(s):
            singletext_result.append(item[0])
        allresult.append(singletext_result)

typetxt.close()
print (u'文字類別析出完畢！結果已輸出到desktop的txttype.txt！')
#直接打印出結果
#for singletext_result in allresult:
#    for item in singletext_result:
#       print item

#所有結果寫入一個txt，一行一個文字
for singletext_result in allresult:
    for item in singletext_result:
        wordseg_result.write(item+'\t')
    wordseg_result.write('\n')
wordseg_result.close()
print (u'分詞完畢！分詞結果已輸出到desktop的wordseg_result.txt！'+'\n')


#過濾停用詞
stopwords = []
delstopwords_alltxt = []

st = codecs.open('C:\\Users\\Administrator\\PycharmProjects\\Newtextmining\\File\\stopwords.txt', 'rb',encoding='utf-8')
delstopwords_result = codecs.open('C:\\Users\\Administrator\\Desktop\\delstopwords_result.txt' , 'a',encoding='utf-8')

for line in st:
    line = line.strip()
    stopwords.append(line)

print (u'正在過濾停用詞......')

for singletext_result in allresult:
    delstopwords_singletxt = []
    for word in singletext_result:
        word = word.strip()
        if word not in stopwords:
            if word >= u'\u4e00' and word <= u'\u9fa5':#判斷是否是漢字
                delstopwords_singletxt.append(word)
    delstopwords_alltxt.append(delstopwords_singletxt)

for delstopwords_singletxt in delstopwords_alltxt:
    for everyword in delstopwords_singletxt:
        delstopwords_result.write(everyword + '\t')
    delstopwords_result.write('\n')
delstopwords_result.close()
print (u'停用詞過濾完畢！已將結果輸出到desktop的delstopwords_result.txt！'+'\n')


#統計絕對詞頻統計TF
getTF_alltxt_dic = {}
getTF_result =  codecs.open('C:\\Users\\Administrator\\Desktop\\getTF_result.txt' , 'a',encoding='utf-8')

print (u'正在統計TF......')

for delstopwords_singletxt in delstopwords_alltxt:
    getTF_singletxt_dic = {}
    for everyword in delstopwords_singletxt:
        everyword = everyword.strip()
        if everyword in getTF_singletxt_dic:
            getTF_singletxt_dic[everyword] += 1
        else:
            getTF_singletxt_dic[everyword] = 1

    getTF_singletxt_dic = sorted(getTF_singletxt_dic.items(), key=lambda d: d[1], reverse=1)


    for a, b in getTF_singletxt_dic:
        if b > 0:
            getTF_result.write(a + '\t' + str(b) + '\t')
    getTF_result.write('\n')

getTF_result.close()
print (u'TF值統計完畢！已將結果輸出到desktop的getTF_result.txt！'+'\n')


#特徵選擇
#計算所有類別DF
alltext = []
allwords = []

delstopwords_result =  codecs.open('C:\\Users\\Administrator\\Desktop\\delstopwords_result.txt' , 'rb',encoding='utf-8')

wordlist = []
for line in delstopwords_result:
    alltext.append(line)
    words = line.strip('\n').split('\t')
    for word in words:
        if word:
            wordlist.append(word)

print u"原始文字詞彙總數：", len(wordlist)
print u"文字個數：", len(alltext)
print u"詞彙種數：", len(set(wordlist))

print ('\n'+u'正在計算所有類別DF......')
word_df = []
for word in set(wordlist):
    count = 0
    for words in alltext:
        if word in words:
            count += 1
    word_df.append([word, str(count)])  # 儲存形式[word，DF]

# 輸出
word_df.sort(key=lambda x: int(x[1]),reverse=True)  # 詞頻從大到小排序
b = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result.txt', "a", encoding="utf-8")
b.truncate()
for item in word_df:
    for word in item:
        b.write(word+ '\t')
    b.write( '\n')
b.close()

b = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result.txt', "rb", encoding="utf-8")
for line in b:
    line = line.split('\t')
    if len(line[0])>1:
        c  = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result1.txt', "a", encoding="utf-8")
        c.write(line[0]+'\t'+line[1])
        c.write('\n')
c.close()

print (u'所有類別DF值統計完畢！已將結果輸出到desktop的DF_allclass_result1.txt！'+'\n')

#特徵選擇
#計算單一類別DF

print (u'正在計算單一類別DF......')

word_df2 = []
d  = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result1.txt', "rb", encoding="utf-8")

for line in d:
    line = line.split()
    word = line[0]

    count_1 = 0
    count_2 = 0
    count_3 = 0
    count_4 = 0
    count_5 = 0
    count_6 = 0
    count_7 = 0
    count_8 = 0
    count_9 = 0

    for words in alltext[0:500]:
        if word in words:
            count_1 += 1

    for words in alltext[500:1000]:
        if word in words:
            count_2 += 1

    for words in alltext[1000:1500]:
        if word in words:
            count_3 += 1

    for words in alltext[1500:2000]:
        if word in words:
            count_4 += 1

    for words in alltext[2000:2500]:
        if word in words:
            count_5 += 1

    for words in alltext[2500:3000]:
        if word in words:
            count_6 += 1

    for words in alltext[3000:3500]:
        if word in words:
            count_7 += 1

    for words in alltext[3500:4000]:
        if line[0] in words:
            count_8 += 1

    for words in alltext[4000:4500]:
        if word in words:
            count_9 += 1

    word_df2.append([word,str(count_1),str(count_2),str(count_3),str(count_4),str(count_5),str(count_6),str(count_7),str(count_8),str(count_9)])  # 儲存形式[word，DF]

d.close()

# 輸出
e = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_singleclass_result.txt', "a", encoding="utf-8")
for item in word_df2:
    for term in item:
        e.write(term+'\t')
    e.write( '\n')
e.close()

print (u'單一類別DF值統計完畢！已將結果輸出到desktop的DF_singleclass_result.txt！'+'\n')

#計算特徵項資訊熵
print (u'正在計算資訊熵......')

IG = []
g  = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result1.txt', "rb", encoding="utf-8")
for line in g:
    line = line.split()
    word = line[0]
    word2 = float(line[1])
    PC = float(500)/float(4500)
    PC_1 = float(PC) * 9
    Entropy = -(float(PC_1)*float(math.log(PC_1,2)))
    PT = float(word2)/float(len(alltext))
    PT_1 = float(1) - PT

    h = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_singleclass_result.txt', "rb", encoding="utf-8")
    lines = h.readline()
    line = lines[:-1]
    line = line.split()
    PCT_evenplus = ( float(line[1]) + float(line[2]) + float(line[3]) + float(line[4]) + float(line[5]) + float(line[6]) + float(line[7]) + float(line[8]) )/float(word2)
    PCT_evenplus_ = float(1) - float(PCT_evenplus)
    E1 = - (float(PCT_evenplus) * float(math.log(PCT_evenplus, 2)))
    E2 = -(float(PCT_evenplus_) * float(math.log(float(PCT_evenplus),2)))
    exEtropy =float(PT) * float(E1) + float(PT_1) * float(E2)

    IG_value = float(Entropy) - float(exEtropy)
    IG.append([word, str(IG_value)])  # 儲存形式[word，IG_value]

IG.sort(key=lambda x: float(x[1]), reverse=True)  # 詞頻從大到小排序
i = codecs.open('C:\\Users\\Administrator\\Desktop\\IG_value.txt', "a", encoding="utf-8")
i.truncate()
for item in IG:
    for word in item:
        i.write(word + '\t')
    i.write('\n')
i.close()

h.close()
g.close()
print (u'資訊增益值統計完畢！已將結果輸出到desktop的IG_value.txt！'+'\n')

print (u'正在選擇特徵詞......')
j = codecs.open('C:\\Users\\Administrator\\Desktop\\IG_value.txt', "rb", encoding="utf-8")
for line in j:
    line = line.split()
    if float(line[1])> -float(10):
        k = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords.txt', "a", encoding="utf-8")
        k.write(line[0])
        k.write('\n')
k.close()
j.close()
print (u'特徵詞選擇完畢！已將結果輸出到desktop的FeatureWords.txt！'+'\n')

f1 = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result1.txt', "rb", encoding="utf-8")
f2 = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords.txt', "rb", encoding="utf-8")
FeatureWords_value = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords_value.txt', "a", encoding="utf-8")

dic = {}
for line in f1:
    line =line.strip('\n').split('\t')
    dic[line[0]]=line[1]
f1.close()

for word in f2:
    if word in dic:
        FeatureWords_value.write(dic[word]+'\t'+dic[word]+'\n')
f2.close()
FeatureWords_value.close()


#文件向量化
print (u'正在進行文字向量化處理......')

f1 = codecs.open('C:\\Users\\Administrator\\Desktop\\DF_allclass_result1.txt', "rb", encoding="utf-8")
f2 = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords.txt', "rb", encoding="utf-8")
FeatureWords_value = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords_value.txt', "a", encoding="utf-8")

allfw=[]
for line in f1:
    dic = {}
    line =line.strip('\n').split('\t')
    dic[line[0]]=line[1]
    allfw.append(dic)
f1.close()

#for dic in allfw:
#   for k,v in dic.iteritems():
#       print k,v

fw=[]
for line in f2:
    line=line.strip('\n').split('\t')
    fw.append(line[0])
f2.close()

#for word in fw:
#   print word

for dic in allfw:
    for word in fw:
        if word in dic:
            for k, v in dic.iteritems():
                FeatureWords_value.write(k+'\t'+v+'\n')
FeatureWords_value.close()


feture_word = []
feture_word_dic = {}
feture_word_dic2 = {}

FeatureWords_value = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords_value.txt', "rb", encoding="utf-8")
for line in FeatureWords_value:
    line = line.split()
    IDF = math.log(4500/float(line[1]),10)
    feture_word.append(line[0])
    feture_word_dic[line[0]] = line[1]
    feture_word_dic2[line[0]] = IDF
FeatureWords_value.close()

getTF_result = codecs.open('C:\\Users\\Administrator\\Desktop\\getTF_result.txt', "rb", encoding="utf-8")
all=[]
for line in getTF_result:
    line = line.strip('\n').split('\t')
    single=[]
    for words in line:
        single.append(words)
    all.append(single)

#for single in all:
    #print single
  #  for word in single:
   #    print word

FeatureWords_value = codecs.open('C:\\Users\\Administrator\\Desktop\\FeatureWords_value.txt', "rb", encoding="utf-8")
alltext_vector = []
for single in all:

#    print tmax
    vector = []
    for word in feture_word:
        if word in single:
            tmax = single[1]
            inde=single.index(word)
            t = single[inde+1]
        else:
            t = 0
#        print t
        tf_idf = (float(t)/float(tmax))*float(feture_word_dic2[word])
        vector.append(tf_idf)
    alltext_vector.append(vector)

for vector in alltext_vector[0:500]:
    vector.append('Economy')
for vector in alltext_vector[500:1000]:
    vector.append('IT')
for vector in alltext_vector[1000:1500]:
    vector.append( 'Health')
for vector in alltext_vector[1500:2000]:
    vector.append('PE')
for vector in alltext_vector[2000:2500]:
    vector.append('Travel')
for vector in alltext_vector[2500:3000]:
    vector.append( 'Education')
for vector in alltext_vector[3000:3500]:
    vector.append('Enployment')
for vector in alltext_vector[3500:4000]:
    vector.append('Culture')
for vector in alltext_vector[4000:4500]:
    vector.append('Military')

#for vector in alltext_vector:
#  print vector
#    for value in vector:
#       print value

data = codecs.open('C:\\Users\\Administrator\\Desktop\\data.arff', "a", encoding="utf-8")
data.truncate()

data.write(u'@relation'+' '+u'sougoucorpus'+'\n\n')
for everyword in feture_word:
    data.write(u'@attribute'+ ' '+ everyword +' '+u'numeric\n')
data.write(u'@attribute type {Economy,IT,Health,PE,Travel,Educaiton,Enployment,Culture,Military}\n\
 
[email protected]\n')
for vector in alltext_vector:
    for value in vector[:-1]:
        data = codecs.open('C:\\Users\\Administrator\\Desktop\\data.arff', "a", encoding="utf-8")
        data.write(str(value) + ',')
    data.write(str(vector[-1]) + '\n')
    data.close()

print (u'文字向量化處理完畢！已將結果輸出到desktop的data.arff！'+'\n')

print (u'文字預處理結束！'+'\n')