Upload 5 files

README.md

@@ -24,16 +24,16 @@ task_ids:
 
 ## Dataset Description
 
-- **Website:** https://zenodo.org/record/4761505
-- **Paper:** [Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? A Comprehensive Assessment for Catalan](https://arxiv.org/abs/2107.07903)
-
-- **Point of Contact:** [Carlos Rodríguez-Penagos]([email protected]) and [Carme Armentano-Oller]([email protected])
+- **Website:** [Zenodo](https://zenodo.org/record/7334110)
+- **Point of Contact:** [Irene Baucells de la Peña]([email protected]), [Carlos Rodríguez-Penagos]([email protected]) and [Carme Armentano-Oller](carme.armentano@bsc.es)
 
 
 
 ### Dataset Summary
 
-TeCla (Text Classification) is a Catalan News corpus for thematic Text Classification tasks. It contains 153.265 articles classified under 30 different categories.
+TeCla (Text Classification) is a Catalan News corpus for thematic multi-class Text Classification tasks. The present version (2.0) contains 113.376 articles classified under a hierarchical class structure consisting of a coarse-grained and a fine-grained class. Each of the 4 coarse-grained classes accept a subset of fine-grained ones, 53 in total. 
+
+The previous version (1.0.1) can still be found at https://zenodo.org/record/4761505
 
 This dataset was developed by [BSC TeMU](https://temu.bsc.es/) as part of [Projecte AINA](https://politiquesdigitals.gencat.cat/ca/economia/catalonia-ai/aina/), to enrich the [Catalan Language Understanding Benchmark (CLUB)](https://club.aina.bsc.es/).
 
@@ -53,17 +53,21 @@ Three json files, one for each split.
 
 ### Data Fields
 
-We used a simple model with the article text and associated labels, without further metadata.
+Each example contains the following 3 fields: 
+* text: the article text (string)
+* label1: the coarse-grained class
+* label2: the fine-grained class
 
 #### Example:
 
 <pre>
-{"version": "1.0",
+{"version": "2.0",
  "data":
    [
     {
-     'sentence': 'L\'editorial valenciana Media Vaca, Premi Nacional a la Millor Tasca Editorial Cultural del 2018. El jurat en destaca la cura "exquisida" del catàleg, la qualitat dels llibres i el "respecte" pels lectors. ACN Madrid.-L\'editorial valenciana Media Vaca ha obtingut el Premi Nacional a la Millor Labor Editorial Cultural corresponent a l\'any 2018 que atorga el Ministeri de Cultura i Esports. El guardó pretén distingir la tasca editorial d\'una persona física o jurídica que hagi destacat per l\'aportació a la vida cultural espanyola. El premi és de caràcter honorífic i no té dotació econòmica. En el cas de Media Vaca, fundada pel valencià Vicente Ferrer i la bilbaïna Begoña Lobo, el jurat n\'ha destacat la cura "exquisida" del catàleg, la qualitat dels llibres i el "respecte" pels lectors i per la resta d\'agents de la cadena del llibre. Media Vaca va publicar els primers llibres el desembre del 1998. El catàleg actual el componen 64 títols dividits en sis col·leccions, que barregen ficció i no ficció. Des del Ministeri de Cultura es destaca que la il·lustració té un pes "fonamental" als productes de l\'editorial i que la majoria de projectes solen partir de propostes literàries i textos preexistents. L\'editorial ha rebut quatre vegades el Bologna Ragazzi Award. És l\'única editorial estatal que ha aconseguit el guardó que atorga la Fira del Llibre per a Nens de Bolonya, la més important del sector.', 
-    'label': 'Lletres'
+     'sentence': "La setena edició del Festival Fantàstik inclourà les cintes 'Matar a dios' i 'Mandy' i un homenatge a 'Mi vecino Totoro'. Es projectaran 22 curtmetratges seleccionats d'entre més de 500 presentats a nivell internacional. El Centre Cultural de Granollers acull del 8 a l'11 de novembre la setena edició del Festival Fantàstik. El certamen, que s'allargarà un dia, arrencarà amb la projecció de la cinta de Caye Casas i Albert Pide 'Matar a Dios'. Els dos directors estaran presents en la inauguració de la cita. A més, els asssitents podran gaudir de 'Mandy', el darrer treball de Nicolas Cage. Altres llargmetratges seleccionats per aquest any són 'Aterrados' (2017), 'Revenge' (2017), 'A Mata Negra' (2018), 'Top Knot Detective' (2018) i 'La Gran Desfeta' (2018). A més, amb motiu del trentè aniversari de la pel·lícula 'El meu veí Totoro' es durà a terme l'exposició dedicada a aquest film '30 anys 30 artistes' comissariada per Jordi Pastor i Reinaldo Pereira. La mostra '30 anys 30 artistes' recull els treballs de trenta artistes d'estils diferents al voltant de la figura de Totoro i el seu director. Es podrà veure durant els dies de festival i es complementarà amb la projecció de la pel·lícula el diumenge 11 de novembre. Al llarg del festival també es projectaran els 22 curtmetratges prèviament seleccionats d'entre més de 500 presentats a nivell internacional. El millor tindrà una dotació de 1000 euros fruit de la unió de forces amb el Mercat Audiovisual de Catalunya.", 
+    'label1': 'Cultura',
+    'label2': 'Cinema'
     },
     ...
   ]
@@ -74,13 +78,15 @@ We used a simple model with the article text and associated labels, without furt
 
 #### Labels
 
-'Societat', 'Política', 'Turisme', 'Salut', 'Economia', 'Successos', 'Partits', 'Educació', 'Policial', 'Medi ambient', 'Parlament', 'Empresa', 'Judicial', 'Unió Europea', 'Comerç', 'Cultura', 'Cinema', 'Govern', 'Lletres', 'Infraestructures', 'Música', 'Festa i cultura popular', 'Teatre', 'Mobilitat', 'Govern espanyol', 'Equipaments i patrimoni', 'Meteorologia', 'Treball', 'Trànsit', 'Món'
+* label1: 'Societat', 'Política', 'Economia', 'Cultura'
+* label2: 'Llengua', 'Infraestructures', 'Arts', 'Parlament', 'Noves tecnologies', 'Castells', 'Successos', 'Empresa', 'Mobilitat', 'Teatre', 'Treball', 'Logística', 'Urbanisme', 'Govern', 'Entitats', 'Finances', 'Govern espanyol', 'Trànsit', 'Indústria', 'Esports', 'Exteriors', 'Medi ambient', 'Habitatge', 'Salut', 'Equipaments i patrimoni', 'Recerca', 'Cooperació', 'Innovació', 'Agroalimentació', 'Policial', 'Serveis Socials', 'Cinema', 'Memòria històrica', 'Turisme', 'Política municipal', 'Comerç', 'Universitats', 'Hisenda', 'Judicial', 'Partits', 'Música', 'Lletres', 'Religió', 'Festa i cultura popular', 'Unió Europea', 'Moda', 'Moviments socials', 'Comptes públics', 'Immigració', 'Educació', 'Gastronomia', 'Meteorologia', 'Energia'
 
 ### Data Splits
 
-* train.json: 110203 article-label pairs
-* dev.json: 13786  article-label pairs
-* test.json: 13786  article-label pairs
+Train, development and test splits were created in a stratified fashion, following a 0.8, 0.05 and 0.15 proportion, respectively. The sizes of each split are the following:
+* train.json: 90700 examples
+* dev.json: 5669 examples
+* test.json: 17007 examples
 
 ## Dataset Creation
 
@@ -96,9 +102,10 @@ The source data are crawled articles from the Catalan News Agency ([Agència Cat
 
 We crawled 219.586 articles from the Catalan News Agency ([Agència Catalana de Notícies; ACN](https://www.acn.cat/)) newswire archive, the latest from October 11, 2020.
 
-We used the "subsection" category as a classification label, after excluding territorial labels (see [territorial_labels.txt](https://huggingface.co/datasets/projecte-aina/tecla/blob/main/territorial_labels.txt) file) and labels with less than 2000 occurrences. 
+From the crawled data, we selected those articles whose 'section' and 'subsection' categories followed the expected codification combinations included in the ACN's style guide and whose 'section' complied the requirements of containing subsections and being thematically founded (in contrast to geographically defined categories such as 'Món' and 'Unió Europea'). The articles originally belonging to the 'Unió Europea' section, which were related to political organisms from the European Union, were included in the 'Política' coarse-grained category (within a fine-grained category named 'Unió Europea') due to its close proximity between some of the original subsections of 'Política' and those of 'Unió Europea', both defined by the specific political organism dealt with in the article.   
+
+The text field in each example is a concatenation of the original title, subtitle and body of the article (before the concatenation, both title and subtitle were added a final dot whenever they lacked one). The preprocessing of the texts was minimal and consisted in the removal of the pattern "ACN {location}.-" included before the body in each text as well as newlines originally used to divide the text in paragraphs.
 
-With this criteria compiled a total of 153.265 articles for this text classification dataset.
 
 #### Who are the source language producers?
 
@@ -108,11 +115,11 @@ The Catalan News Agency ([Agència Catalana de Notícies; ACN](https://www.acn.c
 
 #### Annotation process
 
-We used the "subsection" category as a classification label, after excluding territorial labels (see [territorial_labels.txt](https://huggingface.co/datasets/projecte-aina/tecla/blob/main/territorial_labels.txt) file) and labels with less than 2000 occurrences.
+The crawled data contained the categories' annotations, which were then used to create this dataset with the mentioned criteria.
 
 #### Who are the annotators?
 
-Editorial staff classified the articles under the different thematic sections, and we extracted these from metadata.
+Editorial staff classified the articles under the different thematic sections and subsections, and we extracted these from metadata.
 
 ### Personal and Sensitive Information
 
@@ -136,7 +143,7 @@ We hope this dataset contributes to the development of language models in Catala
 
 ### Dataset Curators
 
-Text Mining Unit (TeMU) at the Barcelona Supercomputing Center (bsc-temu@bsc.es)
+Irene Baucells ([email protected]), Casimiro Pio Carrino ([email protected]), Carlos Rodríguez ([email protected]) and Carme Armentano (carme.armentano@bsc.es), from [BSC-CNS](https://www.bsc.es/).
 
 This work was funded by the [Departament de la Vicepresidència i de Polítiques Digitals i Territori de la Generalitat de Catalunya](https://politiquesdigitals.gencat.cat/ca/inici/index.html#googtrans(ca|en) within the framework of [Projecte AINA](https://politiquesdigitals.gencat.cat/ca/economia/catalonia-ai/aina).
 
@@ -147,34 +154,5 @@ This work is licensed under a <a rel="license" href="https://creativecommons.org
 
 ### Citation Information
 
-```
-
-@inproceedings{armengol-estape-etal-2021-multilingual,
-    title = "Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? {A} Comprehensive Assessment for {C}atalan",
-    author = "Armengol-Estap{\'e}, Jordi  and
-      Carrino, Casimiro Pio  and
-      Rodriguez-Penagos, Carlos  and
-      de Gibert Bonet, Ona  and
-      Armentano-Oller, Carme  and
-      Gonzalez-Agirre, Aitor  and
-      Melero, Maite  and
-      Villegas, Marta",
-    booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
-    month = aug,
-    year = "2021",
-    address = "Online",
-    publisher = "Association for Computational Linguistics",
-    url = "https://aclanthology.org/2021.findings-acl.437",
-    doi = "10.18653/v1/2021.findings-acl.437",
-    pages = "4933--4946",
-}
-
-```
-
-
-[DOI](https://doi.org/10.5281/zenodo.4529183)
-
-
-### Contributions
+[DOI]([https://doi.org/10.5281/zenodo.7334110])
 
-[N/A]

dev.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1222f41063baa63ce8139691ab3e7021c19b8ccab3a6cb60d40c35eb93df81a2
-size 34946188
+oid sha256:e0a9dcf1778992fac105abb1eed1b1e995a29beabd8ecb258a599424833f9e04
+size 14539836

tecla.py

@@ -5,9 +5,9 @@ import datasets
 logger = datasets.logging.get_logger(__name__)
 
 _CITATION = """
-               Carrino, Casimiro Pio, Rodriguez-Penagos, Carlos Gerardo, & Armentano-Oller, Carme. (2021). 
-               TeCla: Text Classification Catalan dataset (Version 1.0) [Data set]. 
-               Zenodo. http://doi.org/10.5281/zenodo.4627198
+               Baucells, Irene, Carrino, Casimiro Pio, Rodriguez-Penagos, Carlos Gerardo, & Armentano-Oller, Carme. (2021).
+               TeCla: Text Classification Catalan dataset (Version 2.0) [Data set].
+               Zenodo. http://doi.org/10.5281/zenodo.7334110
             """
 
 _DESCRIPTION = """
@@ -19,7 +19,7 @@ _DESCRIPTION = """
 _HOMEPAGE = """https://zenodo.org/record/4761505"""
 
 # TODO: upload datasets to github
-_URL = "https://huggingface.co/datasets/bsc/tecla/resolve/main/"
+_URL = "./"
 _TRAINING_FILE = "train.json"
 _DEV_FILE = "dev.json"
 _TEST_FILE = "test.json"
@@ -43,7 +43,7 @@ class tecla(datasets.GeneratorBasedBuilder):
         teclaConfig(
             name="tecla",
             version=datasets.Version("1.0.1"),
-            description="tecla dataset",
+            description="tecla 2.0 dataset",
         ),
     ]
 
@@ -53,28 +53,71 @@ class tecla(datasets.GeneratorBasedBuilder):
             features=datasets.Features(
                 {
                     "text": datasets.Value("string"),
-                    "label": datasets.features.ClassLabel
+                    "label1": datasets.features.ClassLabel
                         (names=
                     [
-                        "Medi ambient",
                         "Societat",
-                        "Policial",
-                        "Judicial",
-                        "Empresa",
-                        "Partits",
                         "Pol\u00edtica",
-                        "Successos",
-                        "Salut",
+                        "Economia",
+                        "Cultura",
+                    ]
+                    ),
+                    "label2": datasets.features.ClassLabel
+                        (names=
+                    [
+                        "Llengua",
                         "Infraestructures",
+                        "Arts",
                         "Parlament",
-                        "M\u00fasica",
-                        "Govern",
-                        "Uni\u00f3 Europea",
-                        "Economia",
+                        "Noves tecnologies",
+                        "Castells",
+                        "Successos",
+                        "Empresa",
                         "Mobilitat",
+                        "Teatre",
                         "Treball",
-                        "Cultura",
-                        "Educaci\u00f3"
+                        "Log\u00edstica",
+                        "Urbanisme",
+                        "Govern",
+                        "Entitats",
+                        "Finances",
+                        "Govern espanyol",
+                        "Tr\u00e0nsit",
+                        "Ind\u00fastria",
+                        "Esports",
+                        "Exteriors",
+                        "Medi ambient",
+                        "Habitatge",
+                        "Salut",
+                        "Equipaments i patrimoni",
+                        "Recerca",
+                        "Cooperaci\u00f3",
+                        "Innovaci\u00f3",
+                        "Agroalimentaci\u00f3",
+                        "Policial",
+                        "Serveis Socials",
+                        "Cinema",
+                        "Mem\u00f2ria hist\u00f2rica",
+                        "Turisme",
+                        "Pol\u00edtica municipal",
+                        "Comer\u00e7",
+                        "Universitats",
+                        "Hisenda",
+                        "Judicial",
+                        "Partits",
+                        "M\u00fasica",
+                        "Lletres",
+                        "Religi\u00f3",
+                        "Festa i cultura popular",
+                        "Uni\u00f3 Europea",
+                        "Moda",
+                        "Moviments socials",
+                        "Comptes p\u00fablics",
+                        "Immigraci\u00f3",
+                        "Educaci\u00f3",
+                        "Gastronomia",
+                        "Meteorologia",
+                        "Energia"
                     ]
                     ),
                 }
@@ -105,8 +148,10 @@ class tecla(datasets.GeneratorBasedBuilder):
             acn_ca = json.load(f)
             for id_, article in enumerate(acn_ca["data"]):
                 text = article["sentence"]
-                label = article["label"]
+                label1 = article["label1"]
+                label2 = article["label2"]
                 yield id_, {
                     "text": text,
-                    "label": label,
+                    "label1": label1,
+                    "label2": label2,
                 }

test.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ecd080814ed5e466f33f5a5d64eb85c94c13efe367897174f4546491cade6df8
-size 34997197
+oid sha256:7954051e42ed7d97fe0ac0977bc624965fd5af77170c44fffaacdfbb49f472e7
+size 42972092

train.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13501fd5cbb51cdd3bcefaae305ef1e6f9e74209cc7f6255275535e6d374b67d
-size 280571534
+oid sha256:4a23c8adb34b5d70b126c1e83d4ea7f998922bd42daa290edb13d657889a5c52
+size 230031411