CRF++地名实体识别（特征为词性和词）

http://x-algo.cn/index.php/2016/02/29/crf-name-entity-recognition/

类似使用CRF实现分词和词性标注，地域识别也是需要生成相应的tag进行标注。这里使用的语料库是1998年1月人民日报语料集。最终学习出来的模型，对复杂的地名识别准确率（F值）非常低，推测是预料中对地名的标注多处是前后矛盾。例如  [华南/ns 地区/n]ns  标为地名实体，但是 东北/f 地区/n 确分开标注，类似错误还有很多。将来有时间可以考虑使用微软的词库 戳我下载-微软词库。

本文还是在人民日报的语料之下，在分完词的粒度BMES标注最后效果如下：

 

1 2 3 4 5 6 7 8 9 10

------ LOC_E ------- [LOC_E] P = 0.832215, R = 0.629442, F-score = 0.716763 ------ LOC_B ------- [LOC_B] P = 0.781022, R = 0.543147, F-score = 0.640719 ------ LOC_S ------- [LOC_S] P = 0.986800, R = 0.994489, F-score = 0.990629 ------ LOC_I ------- [LOC_I] P = 0.736842, R = 0.442105, F-score = 0.552632 ------ All ------- [All] P = 0.975204, R = 0.957399, F-score = 0.966219

由于单字识别F值很高，并且数量多，所以整个识别的效果还是很高。 语料、相关代码下载：[戳我下载]crf++地名实体识别，下面为具体流程。

文章目录 [展开]

生成训练和测试数据

通过一个python脚本按照一定比例生成训练和测试数据，生成过程中按照BMES对语料进行标识，具体规则如下：

通过调用脚本： cat people-daily.txt | python get_ner_loc_train_test_data.py  >log 生成所需要的训练和测试数据，中间过程打印出来很多调试信息，打印到标准输出话费较多时间。具体代码如下（已折叠）：

 

 

 

 

 

 

Python

 

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123

#coding=utf8   import sys home_dir = "./"   def saveDataFile(trainobj,testobj,isTest,word,handle,tag):     if isTest:         saveTrainFile(testobj,word,handle,tag)     else:         saveTrainFile(trainobj,word,handle,tag)   def saveTrainFile(fiobj,word,handle,tag):     if len(word) > 0 and  word != "。" and word != "，":         fiobj.write(word + ' ' + handle  + ' ' +tag +' ')     else:         fiobj.write(' ')   #填充地点标注，非地点的不添加 def fill_local_tag(words, tags):     pos = 0     while True:         print "pos:", pos, " len:", len(words)           if pos == len(words):             print "添加地点tag执行结束"             print tags             break         word = words[pos]         left = word.find("[")         if left == -1 :             print "单个词", word             w,h = word.split("/")             print w,h             if h == "ns": #单个词是地点                 tags[pos] = "LOC_S"             print "本轮tag",tags[pos]             pos += 1         elif left >= 0:             print "发现词组" ,word             search_pos = pos             for word in words[pos+1:]:                 print word                 search_pos += 1                 if word.find("[") >=0:                     print "括号配对异常"                     sys.exit(255)                 if word.find("]") >=0:                     break             if words[search_pos].find("]")  == -1:                 print "括号配对异常，搜索到句尾没有找都另一半括号"                 sys.exit(255)             else:                 #找到另一半，判断原始标注是不是ns，如果是就进行tag标注                 print "match到一个组", words[pos:search_pos+1]                 h = words[search_pos].split("]")[-1] #最后一个词性                 if h == "ns":                     tags[pos] = "LOC_B" #添加首个词                     for p in range(pos + 1,search_pos + 1):                       tags[p] = "LOC_I" #中间词                     tags[search_pos] = "LOC_E" #找到最后一个词                 else:                     p = pos                     for word in words[pos:search_pos+1]:                         print "hhhhhhh", word                         w,h = word.strip("[").split("]")[0].split("/")                         if h == "ns":                             tags[p] = "LOC_S"                         p += 1                     #移动pos             print "本轮添加的tag", tags[pos:search_pos+1]             pos = search_pos + 1       def convertTag():         fiobj    = open( home_dir + 'people-daily.txt','r')     trainobj = open( home_dir +'train.data','w' )     testobj  = open( home_dir  +'test.data','w')       arr = fiobj.readlines()     i = 0     for a in sys.stdin:         i += 1         a = a.strip(' ')         if a=="":continue         words = a.split(" ")         test = False         if i % 5 == 0:             test = True         words = words[1:]         if len(words) == 0: continue           tags = ["O"] * len(words)         fill_local_tag(words, tags)           pos = -1         for word in words:             pos += 1             print "---->", word             word = word.strip(' ')             if len(word) == 0:                 print "Warning 发现空词"                 continue               l1 = word.find('[')             if l1 >=0:                 word = word[l1+1:]               l2 = word.find(']')             if l2 >= 0:                 word = word[:l2]               w,h = word.split('/')                          saveDataFile(trainobj,testobj,test,w,h,tags[pos])         saveDataFile(trainobj, testobj, test,"","","")                  trainobj.flush()     testobj.flush()   if __name__ == '__main__':         convertTag()

模板文件

 

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

#Unigram U01:%x[-1,0] U02:%x[0,0] U03:%x[1,0] U04:%x[2,0] U05:%x[-2,1] U06:%x[-1,1] U07:%x[0,1] U08:%x[1,1] U09:%x[2,1] U0:%x[-2,0] U10:%x[0,0]/%x[0,1] U11:%x[-2,1]%x[-1,1] U18:%x[0,0]/%x[-1,0] U12:%x[0,0]%x[1,0] U13:%x[0,1]%x[-1,0] U14:%x[0,0]%x[1,1] U15:%x[-1,0]%x[-1,1] U16:%x[-1,0]%x[-2,0] U17:%x[-2,0]%x[-2,1] U18:%x[1,0]%x[2,0] U19:%x[-1,0]%x[1,0] U20:%x[1,0]%x[0,1] U22:%x[-2,1]%x[0,1] U23:%x[-1,1]%x[0,1] U24:%x[-1,1]%x[1,1] U25:%x[0,1]%x[1,1] U26:%x[0,1]%x[2,1] U27:%x[1,1]%x[2,1]

开始训练和测试

通过下面命令执行训练和测试过程：

 

1 2	crf_learn -f 4 -p 4 -c 3 template train.data model > train.rst crf_test -m model test.data > test.rst

分类型计算F值

通过执行： python clc.py test.rst 执行脚本，脚本内容如下：

 

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62

#!/usr/bin/python # -*- coding: utf-8 -*-   import sys   god_dic={"LOC_S":0,"LOC_B":0, "LOC_I":0, "LOC_E":0} pre_dic={"LOC_S":0,"LOC_B":0, "LOC_I":0, "LOC_E":0} correct_dic={"LOC_S":0,"LOC_B":0, "LOC_I":0, "LOC_E":0}   if __name__=="__main__":     try:         file = open(sys.argv[1], "r")     except:         print "result file is not specified, or open failed!"         sys.exit()     wc = 0     loc_wc = 0     wc_of_test = 0     wc_of_gold = 0     wc_of_correct = 0     flag = True          for l in file:         wc  += 1         if l==' ': continue         _,_, g, r = l.strip().split()         #并不涉及到地点实体识别         if "LOC" not in g and "LOC" not in r: continue         loc_wc += 1         if "LOC" in g:             god_dic[g]+= 1         if "LOC" in r:             pre_dic[r]+=1         if g == r:             correct_dic[r]+=1           print "WordCount from result:", wc     print "WordCount of loc_wc  post :", loc_wc     print "真实位置标记个数：", god_dic     print "预估位置标记个数：",pre_dic     print "正确标记个数：", correct_dic       res ={"LOC_S":0.0,"LOC_B":0.0, "LOC_I":0.0, "LOC_E":0.0}       all_gold = 0     all_correct = 0     all_pre = 0     for k in god_dic:         print "------ %s -------"%(k)         R = correct_dic[k]/float(god_dic[k])         P = correct_dic[k]/float(pre_dic[k])         print "[%s] P = %f, R = %f, F-score = %f" % (k,P, R, (2*P*R)/(P+R))           all_pre += pre_dic[k]         all_correct += correct_dic[k]         all_gold += god_dic[k]     print "------ All -------"     all_R = all_correct/float(all_gold)     all_P = all_correct/float(all_pre)     print "[%s] P = %f, R = %f, F-score = %f" % ("All",all_P, all_R, (2*all_P*all_R)/(all_P+all_R))

参考文献

基于 CRF和规则相结合的地理命名实体识别方法 何炎祥1,2 罗楚威2 胡彬尧