JorgeAV commited on
Commit
d1c6e47
·
verified ·
1 Parent(s): 0d3f80a

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +84 -24
README.md CHANGED
@@ -17,53 +17,113 @@ task_categories:
17
  - image-to-text
18
  - text-generation
19
  - question-answering
20
- pretty_name: "Expediente 23-F: Documentos Desclasificados del Golpe de Estado"
21
  size_categories:
22
- - n<1K
23
  ---
24
 
25
- # Expediente 23-F: Documentos Desclasificados
26
 
27
- Dataset educativo con documentos desclasificados del intento de golpe de Estado del 23 de febrero de 1981 en Espana (23-F). Cada fila corresponde a una pagina de documento.
28
 
29
- ## Estructura
30
 
31
- | Columna | Tipo | Descripcion |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
32
  |---|---|---|
33
- | `image` | Image | Imagen renderizada de la pagina del documento |
34
- | `pdf_source` | string | Ruta relativa del PDF/imagen original |
35
- | `page_number` | int | Numero de pagina dentro del documento |
36
- | `text` | string | Texto extraido de la pagina (OCR / pdf-parse) |
37
- | `ministry` | string | Ministerio de procedencia (interior, exteriores, defensa) |
38
- | `archive_id` | string | Signatura del archivo (ej: AGA-83-07633, AGMAE-R39017) |
39
 
40
- ## Procedencia de los documentos
41
 
42
- - **interior/** Transcripciones de escuchas telefonicas de la Guardia Civil
43
- - **exteriores/** — Documentos diplomaticos del Ministerio de Asuntos Exteriores
44
- - **defensa/** Documentos del Ministerio de Defensa (informes CESID/CNI)
 
 
 
45
 
46
- ## Contexto historico
47
 
48
- El 23 de febrero de 1981, el teniente coronel Antonio Tejero irrumpio en el Congreso de los Diputados con guardias civiles armados durante la votacion de investidura de Leopoldo Calvo-Sotelo. El golpe fracaso gracias a la intervencion del rey Juan Carlos I.
 
 
 
49
 
50
- Estos documentos fueron desclasificados y publicados por el Gobierno de Espana a traves del portal de Archivos Estatales.
 
51
 
52
- ## Uso
 
 
 
 
 
 
 
 
 
53
 
54
  ```python
55
  from datasets import load_dataset
56
 
57
  ds = load_dataset("JorgeAV/23f-expediente")
58
 
59
- # Ver una pagina
60
  print(ds["train"][0]["text"])
61
  ds["train"][0]["image"].show()
62
 
63
- # Filtrar por ministerio
64
  interior = ds["train"].filter(lambda x: x["ministry"] == "interior")
 
 
 
65
  ```
66
 
67
- ## Licencia
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
68
 
69
- Los documentos originales son de dominio publico (documentos desclasificados del Gobierno de Espana). El dataset estructurado se distribuye bajo CC-BY-4.0.
 
17
  - image-to-text
18
  - text-generation
19
  - question-answering
20
+ pretty_name: "Expediente 23-F: Declassified Documents from the 1981 Spanish Coup Attempt"
21
  size_categories:
22
+ - 1K<n<10K
23
  ---
24
 
25
+ # Expediente 23-F: Declassified Documents from the 1981 Spanish Coup Attempt
26
 
27
+ A page-level multimodal dataset of declassified Spanish government documents related to the failed coup d'etat of February 23, 1981 (known as 23-F). Each row contains a rendered page image paired with its extracted text, sourced from official state archives.
28
 
29
+ ## Historical Context
30
 
31
+ On February 23, 1981, Lieutenant Colonel Antonio Tejero stormed the Spanish Congress of Deputies with armed Guardia Civil officers during the investiture vote of Prime Minister Leopoldo Calvo-Sotelo. The coup failed after King Juan Carlos I publicly opposed it in a televised address. These documents were declassified and published by the Spanish Government through the *Portal de Archivos Estatales*.
32
+
33
+ ## Dataset Schema
34
+
35
+ | Column | Type | Description |
36
+ |---|---|---|
37
+ | `image` | `Image` | Rendered page image (150 DPI) |
38
+ | `pdf_source` | `string` | Relative path to the original PDF or image file |
39
+ | `page_number` | `int32` | Page number within the source document (1-indexed) |
40
+ | `text` | `string` | Extracted text for the page (OCR or pdf-parse) |
41
+ | `ministry` | `string` | Originating ministry (`interior`, `exteriores`, `defensa`) |
42
+ | `archive_id` | `string` | Archival signature (e.g., `AGA-83-07633`, `AGMAE-R39017`) |
43
+
44
+ ## Dataset Statistics
45
+
46
+ | Metric | Value |
47
+ |---|---|
48
+ | **Total rows (pages)** | 1,224 |
49
+ | **Source documents** | 167 |
50
+ | **Text coverage** | 100% (all pages have extracted text) |
51
+ | **Total extracted text** | ~12.6 million characters |
52
+ | **File types** | 155 PDFs (92.8%), 12 JPGs (7.2%) |
53
+
54
+ ### Documents by Ministry
55
+
56
+ | Ministry | Documents | Description |
57
  |---|---|---|
58
+ | `defensa` | 108 | Military intelligence reports (CESID/CNI), internal memos |
59
+ | `exteriores` | 31 | Diplomatic cables, embassy communications, verbal notes |
60
+ | `interior` | 28 | Wiretap transcripts from the Guardia Civil, phone intercepts between conspirators |
 
 
 
61
 
62
+ ### Document Characteristics
63
 
64
+ | Property | Breakdown |
65
+ |---|---|
66
+ | **Format** | 90 typed, 73 scanned, 4 handwritten |
67
+ | **Classification level** | 88 *secreto*, 77 *reservado*, 2 *confidencial* |
68
+ | **Pages per document** | Min: 1, Max: 312, Mean: 7.3 |
69
+ | **Text length per document** | Min: 130 chars, Median: 26.8K chars, Max: 3.45M chars |
70
 
71
+ ## Data Processing Pipeline
72
 
73
+ ### Image Rendering
74
+ - **Tool:** `pdf2image` + Poppler
75
+ - **Resolution:** 150 DPI (PNG)
76
+ - **JPG documents** are loaded directly as single-page images
77
 
78
+ ### Text Extraction
79
+ Two extraction methods were used depending on document type:
80
 
81
+ 1. **pdf-parse** (typed documents): Direct text extraction from digitally-created PDFs
82
+ 2. **Vision-Language OCR** (scanned/handwritten documents):
83
+ - **Model:** [`mlx-community/GLM-OCR-bf16`](https://huggingface.co/mlx-community/GLM-OCR-bf16) running on Apple Silicon via MLX
84
+ - **Input resolution:** 200 DPI
85
+ - **Max tokens per page:** 8,192
86
+ - **Post-processing:** Whitespace normalization, paragraph structure preservation
87
+
88
+ The `isOcrText` field in the source metadata indicates which method was used per document.
89
+
90
+ ## Usage
91
 
92
  ```python
93
  from datasets import load_dataset
94
 
95
  ds = load_dataset("JorgeAV/23f-expediente")
96
 
97
+ # Display a page
98
  print(ds["train"][0]["text"])
99
  ds["train"][0]["image"].show()
100
 
101
+ # Filter by ministry
102
  interior = ds["train"].filter(lambda x: x["ministry"] == "interior")
103
+
104
+ # Get all pages from a specific document
105
+ doc_pages = ds["train"].filter(lambda x: x["pdf_source"] == "interior/guardia-civil/23F_1_conversacion_gc_tejero.pdf")
106
  ```
107
 
108
+ ## Intended Uses
109
+
110
+ - **Historical research** on the Spanish Transition to democracy
111
+ - **Document AI** benchmarking: OCR, layout analysis, and document understanding on Spanish historical documents
112
+ - **NLP in Spanish**: text extraction quality evaluation, named entity recognition on official/archival language
113
+ - **Educational projects** about 20th-century European political history
114
+
115
+ ## Limitations
116
+
117
+ - OCR quality is lower on handwritten documents (4 out of 167)
118
+ - Some scanned pages may contain text artifacts or recognition errors
119
+ - Text extraction from multi-column layouts may not preserve reading order perfectly
120
+ - All text is in Spanish; no translations are provided
121
+
122
+ ## Source & License
123
+
124
+ - **Original documents:** Public domain (declassified records from the Spanish Government, published via the *Portal de Archivos Estatales*)
125
+ - **Structured dataset:** [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)
126
+
127
+ ## Associated Project
128
 
129
+ This dataset powers [23f-expediente](https://github.com/JorgeAV/23f-expediente), an interactive educational web application with an FBI/spy case-file aesthetic for exploring the 23-F documents.