首页 > 代码库 > CRF++地名实体识别(特征为词性和词)

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:

技术分享

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 + ‘\t‘ + handle  + ‘\t‘ +tag +‘\n‘)
    else:
        fiobj.write(‘\n‘)
 
#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(‘\r\n\t ‘)
        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(‘\t ‘)
            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==‘\n‘: 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

CRF++地名实体识别(特征为词性和词)