First version of the pico-breast-cancer dataset.

pico-breast-cancer.py

@@ -0,0 +1,110 @@
+import datasets
+
+
+_CITATION = """\
+@InProceedings{mutinda2022pico,
+  title = {PICO Corpus: A Publicly Available Corpus to Support Automatic Data Extraction from Biomedical Literature},
+  author = {Mutinda, Faith and Liew, Kongmeng and Yada, Shuntaro and Wakamiya, Shoko and Aramaki, Eiji},
+  booktitle = {Proceedings of the first Workshop on Information Extraction from Scientific Publications},
+  pages = {26--31},
+  year = {2022}
+}
+"""
+
+_DESCRIPTION = """\
+The corpus consists of about 1,011 PubMed abstracts which are RCTs related
+to breast cancer. For each abstract, text snippets that identify the
+Participants, Intervention, Control, and Outcome (PICO elements) are annotated.
+The abstracts were annotated using BRAT (https://brat.nlplab.org/) and later
+converted to CoNLL-2003.
+"""
+
+_HOMEPAGE = "https://github.com/Martin-Masson/pico-corpus"
+
+_URL = "https://raw.githubusercontent.com/Martin-Masson/pico-breast-cancer/main/pico_conll.txt"
+
+_TAGS = [
+    "O",
+    "total-participants",
+    "intervention-participants",
+    "control-participants",
+    "age",
+    "eligibility",
+    "ethinicity",
+    "condition",
+    "location",
+    "intervention",
+    "control",
+    "outcome",
+    "outcome-Measure",
+    "iv-bin-abs",
+    "cv-bin-abs",
+    "iv-bin-percent",
+    "cv-bin-percent",
+    "iv-cont-mean",
+    "cv-cont-mean",
+    "iv-cont-median",
+    "cv-cont-median",
+    "iv-cont-sd",
+    "cv-cont-sd",
+    "iv-cont-q1",
+    "cv-cont-q1",
+    "iv-cont-q3",
+    "cv-cont-q3",
+]
+
+
+class PicoBreastCancer(datasets.GeneratorBasedBuilder):
+    """A corpus of about 1,011 PubMed abstracts from RCTs related to breast cancer."""
+
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("string"), 
+                    "tokens": datasets.Sequence(datasets.Value("string")), 
+                    "ner_tags": datasets.Sequence(datasets.ClassLabel(names=_TAGS)),
+                }
+            ),
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_file = dl_manager.download_and_extract(_URL)
+
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.ALL,
+                gen_kwargs={
+                    "filepath": data_file,
+                },
+            )
+        ]
+
+    def _generate_examples(self, filepath):
+        idx = 0
+        tokens = []
+        ner_tags = []
+        labels = {tag: i for i, tag in enumerate(_TAGS)}
+
+        with open(filepath) as f:
+            lines = f.read().splitlines()
+            for line in lines:
+                if not line:
+                    yield id, {
+                        "id": str(idx),
+                        "tokens": tokens,
+                        "ner_tags": ner_tags
+                    }
+                    idx += 1
+                    tokens.clear()
+                    ner_tags.clear()
+                else:
+                    token_tag = line.split()
+                    tokens.append(token_tag[0])
+                    ner_tags.append(labels[token_tag[1]])