First version of SentiNews dataset

dataset_infos.json

@@ -0,0 +1 @@
+{"document_level": {"description": "SentiNews is a Slovenian sentiment classification dataset, consisting of news articles manually annotated with their \nsentiment by between 2 and 6 annotators. The news articles contain political, business, economic and financial content \nfrom the Slovenian news portals 24ur, Dnevnik, Finance, Rtvslo, and \u017durnal24. The texts were annotated using the \nfive-level Lickert scale (1 \u2013 very negative, 2 \u2013 negative, 3 \u2013 neutral, 4 \u2013 positive, and 5 \u2013 very positive) on three \nlevels of granularity, i.e. on the document, paragraph, and sentence level. The final sentiment is determined using \nthe following criterion: negative (if average of scores \u2264 2.4); neutral (if average of scores is between 2.4 and 3.6); \npositive (average of annotated scores \u2265 3.6).\n", "citation": "@article{buvcar2018annotated, \n        title={Annotated news corpora and a lexicon for sentiment analysis in Slovene}, \n        author={Bu{\u000b{c}}ar, Jo{\u000b{z}}e and {\u000b{Z}}nidar{\u000b{s}}i{\u000b{c}}, Martin and Povh, Janez}, \n        journal={Language Resources and Evaluation}, \n        volume={52}, \n        number={3}, \n        pages={895--919}, \n        year={2018}, \n        publisher={Springer}\n}\n", "homepage": "https://github.com/19Joey85/Sentiment-annotated-news-corpus-and-sentiment-lexicon-in-Slovene/", "license": "Creative Commons - Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"nid": {"dtype": "uint16", "id": null, "_type": "Value"}, "content": {"dtype": "string", "id": null, "_type": "Value"}, "sentiment": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "content", "output": "sentiment"}, "task_templates": null, "builder_name": "sentinews", "config_name": "document_level", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 21726849, "num_examples": 10427, "dataset_name": "sentinews"}}, "download_checksums": {"https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1110/SentiNews_document-level.txt": {"num_bytes": 23755522, "checksum": "060fe71d719d79e3bcba335abfc146485fa2f3568b5cc1245bd694834d77190c"}}, "download_size": 23755522, "post_processing_size": null, "dataset_size": 21726849, "size_in_bytes": 45482371}, "paragraph_level": {"description": "SentiNews is a Slovenian sentiment classification dataset, consisting of news articles manually annotated with their \nsentiment by between 2 and 6 annotators. The news articles contain political, business, economic and financial content \nfrom the Slovenian news portals 24ur, Dnevnik, Finance, Rtvslo, and \u017durnal24. The texts were annotated using the \nfive-level Lickert scale (1 \u2013 very negative, 2 \u2013 negative, 3 \u2013 neutral, 4 \u2013 positive, and 5 \u2013 very positive) on three \nlevels of granularity, i.e. on the document, paragraph, and sentence level. The final sentiment is determined using \nthe following criterion: negative (if average of scores \u2264 2.4); neutral (if average of scores is between 2.4 and 3.6); \npositive (average of annotated scores \u2265 3.6).\n", "citation": "@article{buvcar2018annotated, \n        title={Annotated news corpora and a lexicon for sentiment analysis in Slovene}, \n        author={Bu{\u000b{c}}ar, Jo{\u000b{z}}e and {\u000b{Z}}nidar{\u000b{s}}i{\u000b{c}}, Martin and Povh, Janez}, \n        journal={Language Resources and Evaluation}, \n        volume={52}, \n        number={3}, \n        pages={895--919}, \n        year={2018}, \n        publisher={Springer}\n}\n", "homepage": "https://github.com/19Joey85/Sentiment-annotated-news-corpus-and-sentiment-lexicon-in-Slovene/", "license": "Creative Commons - Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"nid": {"dtype": "uint16", "id": null, "_type": "Value"}, "content": {"dtype": "string", "id": null, "_type": "Value"}, "sentiment": {"dtype": "string", "id": null, "_type": "Value"}, "pid": {"dtype": "uint8", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "content", "output": "sentiment"}, "task_templates": null, "builder_name": "sentinews", "config_name": "paragraph_level", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 23107429, "num_examples": 89999, "dataset_name": "sentinews"}}, "download_checksums": {"https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1110/SentiNews_paragraph-level.txt": {"num_bytes": 24342387, "checksum": "c45535325db5a6ef5073d7a594c68e5a5c60983d2e1b277f0dbf76f9975a9f18"}}, "download_size": 24342387, "post_processing_size": null, "dataset_size": 23107429, "size_in_bytes": 47449816}, "sentence_level": {"description": "SentiNews is a Slovenian sentiment classification dataset, consisting of news articles manually annotated with their \nsentiment by between 2 and 6 annotators. The news articles contain political, business, economic and financial content \nfrom the Slovenian news portals 24ur, Dnevnik, Finance, Rtvslo, and \u017durnal24. The texts were annotated using the \nfive-level Lickert scale (1 \u2013 very negative, 2 \u2013 negative, 3 \u2013 neutral, 4 \u2013 positive, and 5 \u2013 very positive) on three \nlevels of granularity, i.e. on the document, paragraph, and sentence level. The final sentiment is determined using \nthe following criterion: negative (if average of scores \u2264 2.4); neutral (if average of scores is between 2.4 and 3.6); \npositive (average of annotated scores \u2265 3.6).\n", "citation": "@article{buvcar2018annotated, \n        title={Annotated news corpora and a lexicon for sentiment analysis in Slovene}, \n        author={Bu{\u000b{c}}ar, Jo{\u000b{z}}e and {\u000b{Z}}nidar{\u000b{s}}i{\u000b{c}}, Martin and Povh, Janez}, \n        journal={Language Resources and Evaluation}, \n        volume={52}, \n        number={3}, \n        pages={895--919}, \n        year={2018}, \n        publisher={Springer}\n}\n", "homepage": "https://github.com/19Joey85/Sentiment-annotated-news-corpus-and-sentiment-lexicon-in-Slovene/", "license": "Creative Commons - Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)", "features": {"nid": {"dtype": "uint16", "id": null, "_type": "Value"}, "content": {"dtype": "string", "id": null, "_type": "Value"}, "sentiment": {"dtype": "string", "id": null, "_type": "Value"}, "pid": {"dtype": "uint8", "id": null, "_type": "Value"}, "sid": {"dtype": "uint8", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "content", "output": "sentiment"}, "task_templates": null, "builder_name": "sentinews", "config_name": "sentence_level", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 24641935, "num_examples": 168899, "dataset_name": "sentinews"}}, "download_checksums": {"https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1110/SentiNews_sentence-level.txt": {"num_bytes": 27220223, "checksum": "4382e064543b7fc6d61cc76bc2f16f8f37c80065dbc4f87b888919d1f45bb9c1"}}, "download_size": 27220223, "post_processing_size": null, "dataset_size": 24641935, "size_in_bytes": 51862158}}

sentinews.py

@@ -0,0 +1,93 @@
+"""SentiNews: Manually sentiment annotated Slovenian news corpus."""
+
+
+import csv
+
+import datasets
+
+_CITATION = """\
+@article{buvcar2018annotated, 
+        title={Annotated news corpora and a lexicon for sentiment analysis in Slovene}, 
+        author={Bu{\v{c}}ar, Jo{\v{z}}e and {\v{Z}}nidar{\v{s}}i{\v{c}}, Martin and Povh, Janez}, 
+        journal={Language Resources and Evaluation}, 
+        volume={52}, 
+        number={3}, 
+        pages={895--919}, 
+        year={2018}, 
+        publisher={Springer}
+}
+"""
+
+_DESCRIPTION = """\
+SentiNews is a Slovenian sentiment classification dataset, consisting of news articles manually annotated with their 
+sentiment by between 2 and 6 annotators. The news articles contain political, business, economic and financial content 
+from the Slovenian news portals 24ur, Dnevnik, Finance, Rtvslo, and Žurnal24. The texts were annotated using the 
+five-level Lickert scale (1 – very negative, 2 – negative, 3 – neutral, 4 – positive, and 5 – very positive) on three 
+levels of granularity, i.e. on the document, paragraph, and sentence level. The final sentiment is determined using 
+the following criterion: negative (if average of scores ≤ 2.4); neutral (if average of scores is between 2.4 and 3.6); 
+positive (average of annotated scores ≥ 3.6).
+"""
+
+_HOMEPAGE = "https://github.com/19Joey85/Sentiment-annotated-news-corpus-and-sentiment-lexicon-in-Slovene/"
+
+_LICENSE = "Creative Commons - Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)"
+
+_URLS = {
+    "document_level": "https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1110/SentiNews_document-level.txt",
+    "paragraph_level": "https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1110/SentiNews_paragraph-level.txt",
+    "sentence_level": "https://www.clarin.si/repository/xmlui/bitstream/handle/11356/1110/SentiNews_sentence-level.txt"
+}
+
+
+class Sentinews(datasets.GeneratorBasedBuilder):
+    """SentiNews: Manually sentiment annotated Slovenian news corpus. Version 1.0."""
+
+    VERSION = datasets.Version("1.0.0")
+
+    BUILDER_CONFIGS = [
+        datasets.BuilderConfig(name="document_level", version=VERSION, description="Dataset annotated at document level."),
+        datasets.BuilderConfig(name="paragraph_level", version=VERSION, description="Dataset annotated at paragraph level."),
+        datasets.BuilderConfig(name="sentence_level", version=VERSION, description="Dataset annotated at sentence level."),
+    ]
+
+    def _info(self):
+        _config_features = {
+            "nid": datasets.Value("uint16"),
+            "content": datasets.Value("string"),
+            "sentiment": datasets.Value("string")
+        }
+
+        if self.config.name == "paragraph_level":
+            _config_features["pid"] = datasets.Value("uint8")
+        elif self.config.name == "sentence_level":
+            _config_features["pid"] = datasets.Value("uint8")
+            _config_features["sid"] = datasets.Value("uint8")
+
+        features = datasets.Features(_config_features)
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            supervised_keys=("content", "sentiment"),
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        urls = _URLS[self.config.name]
+        data_file = dl_manager.download_and_extract(urls)
+        return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_file": data_file})]
+
+    def _generate_examples(self, data_file):
+        _keys_to_return = ["nid", "content", "sentiment"]
+
+        if self.config.name == "paragraph_level":
+            _keys_to_return.append("pid")
+        elif self.config.name == "sentence_level":
+            _keys_to_return.append("pid")
+            _keys_to_return.append("sid")
+
+        with open(data_file, encoding="utf-8") as f:
+            data = csv.DictReader(f, delimiter="\t")
+            for idx, row in enumerate(data):
+                yield idx, {_k: row[_k] for _k in _keys_to_return}