命名体是识别后关系抽取及知识图谱扩充（第一步，数据集构建）_知识图谱结构的扩充

项目场景：

这里我就不介绍碳酸盐台地知识图谱的构建， 如果想看用什么方法构建知识图谱，就看一下前面的博客已经有详细的介绍，这里我就不赘述了。这里还是紧接着用已经完成前面的命名体识别，接着做关系抽取，最终反向补全原有的知识图谱，再经过专家的核对，再将词典补充，重新抽取命名体识别，再再做关系抽取，再反向补全，最最终完善起来。这就是学术中的human in the loop（人在回路），这是整个大类的第二步。

问题描述

目前依旧是那个问题，我缺少数据集，比之前好的是，我有很多已经命名体识别出的文献和中间文本（含有命名体的语句），这我只需要做一些工程，这样就变得很简单。

原因分析：

这里是关系提取需要转变一下思想，我们不需要做具体的词和具体词之间的关系，只需要关键词和关键词之间的关系。换句话说就是我不需要知道是 A Formation 还是 B Formation 还是 AB Formation 对应的某个物质，这是为什么呢，因为我们有具体物质层级关系结构，如图我只要知道Formation 和某个物质父子关系就行，不需要具体名词，在确定关系之后，我们再对这句话进行实体抽取。为什么要这样，而不是实体抽取完了，再关系抽取，这只是对这个项目而言，这样减少数据的冗余，减少运算时间。那下一步就是进行数据的构建

按照上述思路，数据集的构建就很简单，因为只要将词典做简单层级分类就知道谁和谁是父子类关系，父类包含M个实体，子类包含N个实体，那M1对应N3依旧会是父子类关系，因此，我们将这些实体放入固定模板中，那有人会问了，存不存在两个实体多个意思呢，βヾ(,・∇・,川这是不存在的，如果存在就不会放入固定模板这么傻的例子了，只是为了节约时间训练出可用模型，之后等实体关系更多可以扩展。

代码实现：

以时间和物质关系为例，构建模型代码如下：

tim=[]
with open("./duc.txt", "r", encoding="utf-8") as f:
    lines = f.readlines()
    # 去除换行符
    result = ([x.strip() for x in lines if x.strip() != ''])
    # 将整理好字典提取成为全局字典
    for x in result:
        tim.append(x)
sub=[]
with open("./geosubstance1.txt", "r", encoding="utf-8") as f:
    lines = f.readlines()
    # 去除换行符
    result = ([x.strip() for x in lines if x.strip() != ''])
    # 将整理好字典提取成为全局字典
    for x in result:
        sub.append(x)
Result=[]
for Sindex in range(len(sub)):
    for Tindex in range(len(tim)):
        Result.append(sub[Sindex]+' | '+tim[Tindex]+' | unknown | '+tim[Tindex]+' foreland flexure been also accommodated '+sub[Sindex])
for index in range(len(Result)):
    print(Result[index])
f = open('./middle/general.txt', "w", encoding='utf-8')
for line in Result:
    f.write(line + '\n')
print("保存成功")
f.close()
print("okA")
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

最终经过不停的换父子级数据，生成最后的关系抽取数据集部分如下：

Cement | Cretaceous | unknown | Cretaceous foreland flexure been also accommodated Cement
Cement | Berriasian | unknown | Berriasian foreland flexure been also accommodated Cement
Cement | Valanginian | unknown | Valanginian foreland flexure been also accommodated Cement
Cement | Hauterivian | unknown | Hauterivian foreland flexure been also accommodated Cement
Cement | Barremian | unknown | Barremian foreland flexure been also accommodated Cement
Cement | Aptian | unknown | Aptian foreland flexure been also accommodated Cement
Cement | Albian | unknown | Albian foreland flexure been also accommodated Cement
Cement | Cenomanian | unknown | Cenomanian foreland flexure been also accommodated Cement
Cement | Turonian | unknown | Turonian foreland flexure been also accommodated Cement
Cement | Coniacian | unknown | Coniacian foreland flexure been also accommodated Cement
Cement | Santonian | unknown | Santonian foreland flexure been also accommodated Cement
Cement | Campanian | unknown | Campanian foreland flexure been also accommodated Cement
Cement | Maastrichtian | unknown | Maastrichtian foreland flexure been also accommodated Cement
Acicular | Cretaceous | unknown | Cretaceous foreland flexure been also accommodated Acicular
Acicular | Berriasian | unknown | Berriasian foreland flexure been also accommodated Acicular
Acicular | Valanginian | unknown | Valanginian foreland flexure been also accommodated Acicular
Acicular | Hauterivian | unknown | Hauterivian foreland flexure been also accommodated Acicular
Acicular | Barremian | unknown | Barremian foreland flexure been also accommodated Acicular
Acicular | Aptian | unknown | Aptian foreland flexure been also accommodated Acicular
Acicular | Albian | unknown | Albian foreland flexure been also accommodated Acicular
Acicular | Cenomanian | unknown | Cenomanian foreland flexure been also accommodated Acicular
Acicular | Turonian | unknown | Turonian foreland flexure been also accommodated Acicular

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

好了第一步已经完成，后面就是代码训练，在下一篇博客中代码数据会一起上传到github中。