57 lines
1.1 KiB
Markdown
57 lines
1.1 KiB
Markdown
---
|
|
title: Cesar Abascal Gutierrez
|
|
published: true
|
|
taxonomy:
|
|
category: [iaeste]
|
|
tag: [ner,nlp]
|
|
author: Daniel Hladek
|
|
---
|
|
|
|
## Named entity annotations
|
|
|
|
Intern, probably summer 2019
|
|
|
|
Cesar Abascal Gutierrez <cesarbielva1994@gmail.com>
|
|
|
|
## Goals
|
|
|
|
- Be able to recognize unknown named entities
|
|
- Create a manually annotated training set from speech transcripts
|
|
- Propose an annotation schema
|
|
|
|
|
|
## Plan
|
|
|
|
- Convert speech transcripts into a training set
|
|
- Train and evaluate classifier
|
|
- Establish manual annotation
|
|
- Select unannotated data
|
|
|
|
### Data preparation
|
|
|
|
Input: Transcriber transcripts with inconsistent annotations
|
|
|
|
```
|
|
* First small letter: regular word
|
|
* Capital: named entity
|
|
* ''^^'': faoreign word
|
|
* ''@'': noise
|
|
* ''_'': multi word expression
|
|
* ''/'': pronuncation
|
|
```
|
|
|
|
Output: A file that can be read by `spacy convert`
|
|
|
|
## People
|
|
|
|
- Cesar Abascal Gutierrez <cesarbielva1994@gmail.com>
|
|
- Kyryl Kobzar
|
|
- Ediz Morochovič
|
|
|
|
## Tools
|
|
|
|
```
|
|
* Machine learning : https://spacy.io/usage/training
|
|
* Manual Annotation : https://prodi.gy/
|
|
```
|