> For the complete documentation index, see [llms.txt](https://anxiang1836.gitbook.io/nlp-keypoints/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://anxiang1836.gitbook.io/nlp-keypoints/zhong-wen-ner-le-ying/01-latticelstm.md).

# LatticeLSTM

> LatticeLSTM是"词汇增强"方法的中文NER的开篇之作，提出了一种Lattice LSTM以融合词汇信息。
>
> **paper :** [Chinese NER Using Lattice LSTM](https://arxiv.org/pdf/1805.02023.pdf)
>
> **source:** ACL-2018
>
> **code:** [Lattice-LSTM](https://github.com/jiesutd/LatticeLSTM)

主要的核心思想：**将潜在词信息整合到基于字符的 LSTM-CRF 中**

由于在网格中存在指数级数量的词-字符路径，因此研究者利用 lattice LSTM 结构自动控制从句子开头到结尾的信息流。

如下图所示，门控单元用于将来自不同路径的信息动态传送到每个字符，不会受到分词偏差的影响。

![](https://pictrue-bed.oss-cn-beijing.aliyuncs.com/20220911224720.png)

## 1: 整体模型构成

当我们通过词汇信息（词典）匹配一个句子时，可以获得一个类似Lattice的结构。

Lattice是一个有向无环图，词汇的开始和结束字符决定了其位置。Lattice LSTM结构则融合了词汇信息到原生的LSTM中。

![](https://pictrue-bed.oss-cn-beijing.aliyuncs.com/20220911231755.png)

Lattice LSTM引入了一个word cell结构，对于当前的字符，融合以该字符结束的所有word信息，如对于「店」融合了「人和药店」和「药店」的信息。

Lattice LSTM在基于字符的模型上附加了基于词汇对的cells和控制信息流的附加门。

## 2: 特点总结

* 优点

Lattice LSTM 的提出，将词汇信息引入，有效提升了NER性能；

* 缺点

计算性能低下，不能batch并行化：

> 究其原因主要是每个字符之间的增加word cell（看作节点）数目不一致；

信息损失：

> 1）每个字符只能获取以它为结尾的词汇信息，对于其之前的词汇信息也没有持续记忆。
>
> 2）由于RNN特性，采取BiLSTM时其前向和后向的词汇信息不能共享。

可迁移性差：

> 只适配于LSTM，不具备向其他网络迁移的特性


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://anxiang1836.gitbook.io/nlp-keypoints/zhong-wen-ner-le-ying/01-latticelstm.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.