条件随机场（CRF）识别命名实体-优快云博客

本文档记录了一次使用 CRF++ 进行命名实体识别的实验过程，包括实验环境配置、使用的工具介绍、特征模板设定及结果评估方法。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

<a href="http://download.youkuaiyun.com/source/1507312">资实验相关资料下载</a>
CRF++使用见<a href="http://blog.youkuaiyun.com/Felomeng/archive/2009/06/22/4288492.aspx" target="_blank">《CRF++的简单使用》</a>
一、实验环境
a) 软件：windows XP pro sp3，visual studio 2008 & Dotnet2.0， CRF++， perl
b) 硬件： CPU: cm420，内存：2G ddr533，　160G 8M sata 富士通
二、实验过程
下面未经特别说明，都是按照作业要求将训练语料分成7:3进行训练和评测所得的结果。
<a name="_Ref233259982">a) 直接应用CRF</a>
 i. 所给定的语料格式非常符合条件随机场的要求，故直接使用条件随机场进行训练测试。（本次试验的文件在包test1.rar中）
1. 转换文档编码为UTF8（CRF++在使用UTF16时会报错）
2. 制定模板，如下：
#Unigram
U00:%x[-2,0]
U01:%x[-1,0]
U02:%x[0,0]
U03:%x[1,0]
U04:%x[2,0]
U10:%x[-1,0]/%x[0,0]
U11:%x[0,0]/%x[1,0]
3. 使用CRF++学习特征（相关信息如下）
a) 命令：crf_learn template_file train_file model
其中template_file是模板文件，train_file是训练语料，都需要事先准备好；model是CRF++根据模板和训练语料生成的文件，用于解码。
 i. template_file文件
1. 模板的基本格式为%x[row,col]，它用于确定输入数据中的一个token
其中，row确定与当前的token的相对行数。col用于确定绝对列数。（如下图：）
<table class="MsoTableGrid" style="margin: auto auto auto 126pt; border-collapse: collapse;" border="1" cellspacing="0" cellpadding="0"><tbody>
<tr style="">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 53.1pt; padding-right: 5.4pt; padding-top: 0cm; border: black 1pt solid;" width="71" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 59pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
col 0
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 62.8pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="84" valign="top">
col 1
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 66.25pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="88" valign="top">
col 2
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 58.95pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 53.1pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="71" valign="top">
row -2
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 59pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
疆
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 62.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="84" valign="top">
Ens
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 66.25pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="88" valign="top">
I-LOC
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 58.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 53.1pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="71" valign="top">
row -1
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 59pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
总
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 62.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="84" valign="top">
Bn
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 66.25pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="88" valign="top">
N
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 58.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 53.1pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="71" valign="top">
row 0
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 59pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
统
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 62.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="84" valign="top">
En
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 66.25pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="88" valign="top">
N
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 58.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
当前行
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 53.1pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="71" valign="top">
row 1
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 59pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
阿
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 62.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="84" valign="top">
Bns
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 66.25pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="88" valign="top">
B-PER
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 58.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 53.1pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="71" valign="top">
row 2
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 59pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">
利
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 62.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="84" valign="top">
Mns
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 66.25pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="88" valign="top">
I-PER
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 58.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="79" valign="top">

</td>
</tr>
</tbody></table>

<table class="MsoTableGrid" style="margin: auto auto auto 126pt; border-collapse: collapse;" border="1" cellspacing="0" cellpadding="0"><tbody>
<tr style="">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; padding-top: 0cm; border: black 1pt solid;" width="200" valign="top">
模板
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
指代的特征
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U00:%x[-2,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
疆
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U01:%x[-1,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
总
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U02:%x[0,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
统
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U03:%x[1,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
阿
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U04:%x[2,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
利
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U10:%x[-1,0]/%x[0,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
总/统
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
U11:%x[0,0]/%x[1,0]
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 150.05pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="200" valign="top">
统/阿
</td>
</tr>
</tbody></table>

2. 特征模板的类型
a) 第一种以字母U开头，为Unigram template。当模板前加上U之后，CRF会自动生成一个特征函数集合。
一个模型生成的特征函数的个数总数为L*N，其中L是输出的类别数，N是根据给定的template扩展出的独立串(unique string )的数目。
b) 第二种特征模板以B开头，即Bigram template
它用于描述Bigram特征。系统将自动产生当前输出token与前一个输出token的组合。产生的可区分的特征的总数是L*L*N，其中L是输出类别数，N是这个模板产生的unique features数。
c) 两种模板的区别
注意：Unigram/Bigram是指输出token的Unigram/Bigrams，而不是特征！
unigram：|output tag|×|从模板中扩展的所有可能串|
bigram: |output tag| × |output tag| × |从模板中扩展的所有可能串|
b) iter=88 terr=0.01365 serr=0.23876 obj=67066.17413 diff=0.00006
其中：iter是迭代次数；terr是词错误率；serr是句错误率；obj是当前对象值，当它收敛时，迭代结束；diff是与上一对象的差。
4. Done!2706.41 s，用时间2706.41s（在电脑1上）。
5. 对测试语料进行测试
a) 命令：crf_test -m model_file test_file > result_file
其中 model_file是刚才生成的model文件，test_file是待测试语料，“>result_file”是重定向语句，指将屏幕输出直接输出到文件result_file中。
b) CRF++的解码速度是很快的，尤其是直接写入文件时。但是因为特征选取的问题，正确率、召回率都不高。
c) 结果使用conlleval.pl程序测评。（其代码在提交包根目录中）
测评的命令为：perl conlleval.pl < output.txt，其中output.txt为待评测文件，需要perl解释器支持。详细结果如下：
<table class="MsoNormalTable" style="margin: auto auto auto 4.9pt; width: 651px; border-collapse: collapse;" border="0" cellspacing="0" cellpadding="0"><tbody>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
LOC:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
63.67%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
72.93%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
67.98
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
5623
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
382251.5
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
ORG:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
21.26%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
35.90%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
26.71
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
4491
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
119954.6
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
PER:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
65.90%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
65.06%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
65.47
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
2554
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
167210.4
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
宏平均
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
53.38667
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
微平均：
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
52.84311
</td>
</tr>
</tbody></table>

 ii. 因为刚才特征选取地特别少，故猜想多加入有效特征可以提高结果，于是把模板定义如下：（本次试验的相关数据文件在包test2.rar中）
1. 模板2：
#Unigram
U00:%x[-2,0]
U01:%x[-1,0]
U02:%x[0,0]
U03:%x[1,0]
U04:%x[2,0]
U5:%x[-2,0]/%x[-1,0]
U6:%x[-1,0]/%x[0,0]
U7:%x[0,0]/%x[1,0]
U8:%x[1,0]/%x[2,0]
2. 相关的实验数据如下：
a) 训练过程：
iter=94 terr=0.00571 serr=0.12313 obj=53321.45523 diff=0.00000
Done!2915.53 s
b) 测试结果：
<table class="MsoNormalTable" style="margin: auto auto auto 4.9pt; width: 651px; border-collapse: collapse;" border="0" cellspacing="0" cellpadding="0"><tbody>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
LOC:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
66.86%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
74.31%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
70.39
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
5456
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
384047.8
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
ORG:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
26.95%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
41.02%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
32.53
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
4048
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
131681.4
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
PER:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
68.29%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
65.67%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
66.96
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
2488
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
166596.5
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
宏平均
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
56.62667
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
微平均：
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
56.89841
</td>
</tr>
</tbody></table>
的确有所进步，但是还是明显显低。
<a name="_Ref233259991">a) 制定规则，改进结果</a>
 i. 对结果进行分析（详见各包中以error开头的文件），可以发现错误主要有以下几种：
1. 同一实体内不同字间的类型不同，则以字类数较多者为准
a) 个数相同时，多数情况下为LOC
2. 实体开头的字必定为B-???格式
3. 实体的开始和结尾都有特定的特征可以遵循（如停用词、动词等作为分界等）
4. 固定实体后跟实体应为B-???格式（如省名后）
5. 实体间间隔较小时可能合并为同一实体
6. ……
 ii. 根据以上特点对结果进行优化，计划依次试验各个规则。但因时间因素，只检测了四五种，其中较有效果的是前两种（即规则1和2），两者结合可以把结果成绩提高12%左右。在test2的结果上加以更正，得到的结果如下：
<table class="MsoNormalTable" style="margin: auto auto auto 4.9pt; width: 651px; border-collapse: collapse;" border="0" cellspacing="0" cellpadding="0"><tbody>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
LOC:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
79.40%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
76.43%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
77.89
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
4966
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
386801.7
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
ORG:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
53.86%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
52.63%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
53.24
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
3457
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
184050.7
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
PER:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
80.88%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
67.09%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
73.34
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
2327
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
170662.2
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
宏平均
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
68.15667
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
微平均：
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
68.9781
</td>
</tr>
</tbody></table>
虽然F值有很大提高，但是还是太不理想
<a name="_Ref233259997">c) 先分词并标注词性信息，再用CRF</a>学习规则
 i. 看来单从字的角度着眼已然不够，于是试图利用分词和词性标注信息。因为题目未给出相应信息，故用分词标注信息先进行分词标注（分词标注工具见附件包根目录）。
 ii. 分词标注后，字的特征如下所示：
<table class="MsoTableGrid" style="margin: auto auto auto 63pt; border-collapse: collapse;" border="1" cellspacing="0" cellpadding="0"><tbody>
<tr style="">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.75pt; padding-right: 5.4pt; padding-top: 0cm; border: black 1pt solid;" width="121" valign="top">
字
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
词性及分词标记
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
实体标记
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
：
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
Sw
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
N
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
印
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
Bns
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
B-LOC
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
度
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
Ens
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
I-LOC
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
首
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
Bd
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
N
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
先
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
Ed
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 90.8pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="121" valign="top">
N
</td>
</tr>
</tbody></table>
 iii. 于是针对其建立模板：
 iv. 以此模板进行训练，得到模型后进行测试，最后用conlleval测得结果如下： 
iter=226 terr=0.00935 serr=0.17661 act=2913330 obj=42785.69115 diff=0.00009
Done!4502.97 s
<table class="MsoNormalTable" style="margin: auto auto auto 4.9pt; width: 651px; border-collapse: collapse;" border="0" cellspacing="0" cellpadding="0"><tbody>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
LOC:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
82.05%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
89.97%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
85.83
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
20309
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
1743121
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
ORG:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
48.36%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
65.12%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
55.5
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
13818
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
766899
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
PER:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
91.52%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
93.15%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
92.33
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
9189
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
848420.4
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
宏平均
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
77.88667
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
微平均：
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
77.53349
</td>
</tr>
</tbody></table>
 v. 对此结果再以用前面建立的规则优化，最终得到结果如下：
<table class="MsoNormalTable" style="margin: auto auto auto 4.9pt; width: 651px; border-collapse: collapse;" border="0" cellspacing="0" cellpadding="0"><tbody>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
LOC:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
90.34%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
90.37%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
90.36
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
18878
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
1705816
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
ORG:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
70.47%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
71.54%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
71
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
12474
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
885654
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
PER:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75">
precision:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
94.85%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
recall:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
92.70%;
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
FB1:
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
93.76
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
8954
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
839527
</td>
</tr>
<tr style="height: 13.5pt;">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 56pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="75"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72"></td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
宏平均
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
85.04
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
微平均：
</td>
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 54pt; padding-right: 5.4pt; height: 13.5pt; padding-top: 0cm; border: #ece9d8;" width="72">
85.12373
</td>
</tr>
</tbody></table>
在此基础上对Test_utf16.ner进行训练，最终得到finalAnswer.txt
三<a name="_Ref233260008">、实验结果对照</a>表
<table class="MsoTableGrid" style="" border="1" cellspacing="0" cellpadding="0"><tbody>
<tr style="">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; padding-top: 0cm; border: black 1pt solid;" width="47" valign="top">
编号
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">
使用策略
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
结果
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">
方法改进
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
性能提升
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="47" valign="top">
1
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">
单字CRF(1)
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约53%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="47" valign="top">
2
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">
单字CRF(1)
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约56.7%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">
使用更多的特征信息
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约3.7%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">
特征对于结果有较大影响，但因硬件条件和时间原因未能引入更多的特征加以佐证。
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="47" valign="top">
3
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">
单字CRF+规则
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约68.5%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">
人工添加规则，对结果进行优化
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约11.8%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">
规则可以弥补机器学习方法的不足，依次（并改变规则的顺序）尝试各种规则。
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="47" valign="top">
4
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">
分词+词性标注+CRF
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约77.7%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">
采用了不同方法
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约9.2%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">
引入词的概念显然
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="47" valign="top">
5
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">
分词+词性标注+CRF+规则
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约85.1%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">
在4基础上引入规则
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">
约7.4%
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">
机器学习方法的某些弊端不随条件的变化而变化
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 35.4pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="47" valign="top">
6
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 70.95pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="95" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 99.2pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="132" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 2cm; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="76" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 205.55pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="274" valign="top">

</td>
</tr>
</tbody></table>
四<a name="_Ref233260012">、未来的工作</a>
a) 尝试更多的规则，尽量减少机器学习方法的弊端；
b) 尝试把分词和词性信息作为不同的属性，看看对结果有什么影响；
c) 改进分词及词性标注的正确率，以便收到更好的命名实体识别的效果。
<a name="_Ref233260017">五、注意事项</a>
a) 编码格式可能造成某些文件无法正常处理，当出现格式错误时要留心一下；
b) 各个程序所需要的分隔符不尽相同，主要是空格和制表符，在遇到问题时注意看是不是分隔符不符合程序要求；
c) 实验过程中开发的一些实用小工具并未提供说明书，但这些小工具界面简洁，使用方便，应该很容易掌握。
<table class="MsoTableGrid" style="margin: auto auto auto 42pt; border-collapse: collapse;" border="1" cellspacing="0" cellpadding="0"><tbody>
<tr style="">
<td style="padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 198.75pt; padding-right: 5.4pt; padding-top: 0cm; border: black 1pt solid;" width="265" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 185.35pt; padding-right: 5.4pt; border-top: black 1pt solid; border-right: black 1pt solid; padding-top: 0cm;" width="247" valign="top">

</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 198.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="265" valign="top">
Felomeng.BackFormation
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 185.35pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="247" valign="top">
用于在标准格式和分词标注格式之间转换，还附带将两种标记合并、将分词标注信息删除两个功能
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 198.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="265" valign="top">
Felomeng.ErrorExtractor
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 185.35pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="247" valign="top">
错误提取工具，可以方便地从结果（带答案）中提取错误，以便于实验分析
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 198.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="265" valign="top">
Felomeng.NERRules
</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 185.35pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="247" valign="top">
本来有四个功能，因为实验中验证了前三个功能效果不佳，固主要功能就是改善结果（对机器学习方法的结果进行规则化改进）。
</td>
</tr>
<tr style="">
<td style="border-bottom: black 1pt solid; border-left: black 1pt solid; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 198.75pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="265" valign="top">

</td>
<td style="border-bottom: black 1pt solid; border-left: #ece9d8; padding-bottom: 0cm; background-color: transparent; padding-left: 5.4pt; width: 185.35pt; padding-right: 5.4pt; border-top: #ece9d8; border-right: black 1pt solid; padding-top: 0cm;" width="247" valign="top">

</td>
</tr>
</tbody></table>

后记：其实结果和使用的训练测试数据的选择很有关系，本人采用的是前70%训练，后30%测试。后经改进选取方法，正确率可以达92%以上，有兴趣的可以改变一下训练语料和测试语料的提取方式。