diff --git a/bigbio/hub/hub_repos/s800/README.md b/bigbio/hub/hub_repos/s800/README.md new file mode 100644 index 00000000..af7b69df --- /dev/null +++ b/bigbio/hub/hub_repos/s800/README.md @@ -0,0 +1,50 @@ +--- +language: + - en +bigbio_language: + - English +license: other +bigbio_license_shortname: other +multilinguality: monolingual +pretty_name: S800 +homepage: https://species.jensenlab.org/ +bigbio_pubmed: true +bigbio_public: true +bigbio_tasks: + - NAMED_ENTITY_RECOGNITION + - NAMED_ENTITY_DISAMBIGUATION +--- + + +# Dataset Card for S800 + +## Dataset Description + +- **Homepage:** https://species.jensenlab.org/ +- **Pubmed:** True +- **Public:** True +- **Tasks:** NER, NED + +S800 comprises 800 PubMed abstracts in which organism mentions were identified and mapped to the corresponding NCBI Taxonomy identifiers. + +To increase the corpus taxonomic mention diversity the S800 abstracts were collected by selecting 100 abstracts from the following 8 categories: bacteriology, botany, entomology, medicine, mycology, protistology, virology and zoology. +S800 has been annotated with a focus at the species level; however, higher taxa mentions (such as genera, families and orders) have also been considered. + + +## Citation Information + +``` +@article{, + title = {The SPECIES and ORGANISMS Resources for Fast and Accurate Identification of Taxonomic Names in Text}, + author = {Pafilis, Evangelos AND Frankild, Sune P. AND Fanini, Lucia AND Faulwetter, Sarah AND Pavloudi, Christina AND Vasileiadou, Aikaterini AND Arvanitidis, Christos AND Jensen, Lars Juhl}, + journal = {PLOS ONE}, + publisher = {Public Library of Science}, + year = {2013}, + month = {06}, + volume = {8}, + pages = {1-6}, + number = {6}, + url = {https://doi.org/10.1371/journal.pone.0065390}, + doi = {10.1371/journal.pone.0065390}, +} +``` diff --git a/bigbio/hub/hub_repos/s800/bigbiohub.py b/bigbio/hub/hub_repos/s800/bigbiohub.py new file mode 100644 index 00000000..f4da7bb7 --- /dev/null +++ b/bigbio/hub/hub_repos/s800/bigbiohub.py @@ -0,0 +1,590 @@ +from collections import defaultdict +from dataclasses import dataclass +from enum import Enum +import logging +from pathlib import Path +from types import SimpleNamespace +from typing import TYPE_CHECKING, Dict, Iterable, List, Tuple + +import datasets + +if TYPE_CHECKING: + import bioc + +logger = logging.getLogger(__name__) + + +BigBioValues = SimpleNamespace(NULL="") + + +@dataclass +class BigBioConfig(datasets.BuilderConfig): + """BuilderConfig for BigBio.""" + + name: str = None + version: datasets.Version = None + description: str = None + schema: str = None + subset_id: str = None + + +class Tasks(Enum): + NAMED_ENTITY_RECOGNITION = "NER" + NAMED_ENTITY_DISAMBIGUATION = "NED" + EVENT_EXTRACTION = "EE" + RELATION_EXTRACTION = "RE" + COREFERENCE_RESOLUTION = "COREF" + QUESTION_ANSWERING = "QA" + TEXTUAL_ENTAILMENT = "TE" + SEMANTIC_SIMILARITY = "STS" + TEXT_PAIRS_CLASSIFICATION = "TXT2CLASS" + PARAPHRASING = "PARA" + TRANSLATION = "TRANSL" + SUMMARIZATION = "SUM" + TEXT_CLASSIFICATION = "TXTCLASS" + + +entailment_features = datasets.Features( + { + "id": datasets.Value("string"), + "premise": datasets.Value("string"), + "hypothesis": datasets.Value("string"), + "label": datasets.Value("string"), + } +) + +pairs_features = datasets.Features( + { + "id": datasets.Value("string"), + "document_id": datasets.Value("string"), + "text_1": datasets.Value("string"), + "text_2": datasets.Value("string"), + "label": datasets.Value("string"), + } +) + +qa_features = datasets.Features( + { + "id": datasets.Value("string"), + "question_id": datasets.Value("string"), + "document_id": datasets.Value("string"), + "question": datasets.Value("string"), + "type": datasets.Value("string"), + "choices": [datasets.Value("string")], + "context": datasets.Value("string"), + "answer": datasets.Sequence(datasets.Value("string")), + } +) + +text_features = datasets.Features( + { + "id": datasets.Value("string"), + "document_id": datasets.Value("string"), + "text": datasets.Value("string"), + "labels": [datasets.Value("string")], + } +) + +text2text_features = datasets.Features( + { + "id": datasets.Value("string"), + "document_id": datasets.Value("string"), + "text_1": datasets.Value("string"), + "text_2": datasets.Value("string"), + "text_1_name": datasets.Value("string"), + "text_2_name": datasets.Value("string"), + } +) + +kb_features = datasets.Features( + { + "id": datasets.Value("string"), + "document_id": datasets.Value("string"), + "passages": [ + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "text": datasets.Sequence(datasets.Value("string")), + "offsets": datasets.Sequence([datasets.Value("int32")]), + } + ], + "entities": [ + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "text": datasets.Sequence(datasets.Value("string")), + "offsets": datasets.Sequence([datasets.Value("int32")]), + "normalized": [ + { + "db_name": datasets.Value("string"), + "db_id": datasets.Value("string"), + } + ], + } + ], + "events": [ + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + # refers to the text_bound_annotation of the trigger + "trigger": { + "text": datasets.Sequence(datasets.Value("string")), + "offsets": datasets.Sequence([datasets.Value("int32")]), + }, + "arguments": [ + { + "role": datasets.Value("string"), + "ref_id": datasets.Value("string"), + } + ], + } + ], + "coreferences": [ + { + "id": datasets.Value("string"), + "entity_ids": datasets.Sequence(datasets.Value("string")), + } + ], + "relations": [ + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "arg1_id": datasets.Value("string"), + "arg2_id": datasets.Value("string"), + "normalized": [ + { + "db_name": datasets.Value("string"), + "db_id": datasets.Value("string"), + } + ], + } + ], + } +) + + +TASK_TO_SCHEMA = { + Tasks.NAMED_ENTITY_RECOGNITION.name: "KB", + Tasks.NAMED_ENTITY_DISAMBIGUATION.name: "KB", + Tasks.EVENT_EXTRACTION.name: "KB", + Tasks.RELATION_EXTRACTION.name: "KB", + Tasks.COREFERENCE_RESOLUTION.name: "KB", + Tasks.QUESTION_ANSWERING.name: "QA", + Tasks.TEXTUAL_ENTAILMENT.name: "TE", + Tasks.SEMANTIC_SIMILARITY.name: "PAIRS", + Tasks.TEXT_PAIRS_CLASSIFICATION.name: "PAIRS", + Tasks.PARAPHRASING.name: "T2T", + Tasks.TRANSLATION.name: "T2T", + Tasks.SUMMARIZATION.name: "T2T", + Tasks.TEXT_CLASSIFICATION.name: "TEXT", +} + +SCHEMA_TO_TASKS = defaultdict(set) +for task, schema in TASK_TO_SCHEMA.items(): + SCHEMA_TO_TASKS[schema].add(task) +SCHEMA_TO_TASKS = dict(SCHEMA_TO_TASKS) + +VALID_TASKS = set(TASK_TO_SCHEMA.keys()) +VALID_SCHEMAS = set(TASK_TO_SCHEMA.values()) + +SCHEMA_TO_FEATURES = { + "KB": kb_features, + "QA": qa_features, + "TE": entailment_features, + "T2T": text2text_features, + "TEXT": text_features, + "PAIRS": pairs_features, +} + + +def get_texts_and_offsets_from_bioc_ann(ann: "bioc.BioCAnnotation") -> Tuple: + + offsets = [(loc.offset, loc.offset + loc.length) for loc in ann.locations] + + text = ann.text + + if len(offsets) > 1: + i = 0 + texts = [] + for start, end in offsets: + chunk_len = end - start + texts.append(text[i : chunk_len + i]) + i += chunk_len + while i < len(text) and text[i] == " ": + i += 1 + else: + texts = [text] + + return offsets, texts + + +def remove_prefix(a: str, prefix: str) -> str: + if a.startswith(prefix): + a = a[len(prefix) :] + return a + + +def parse_brat_file( + txt_file: Path, + annotation_file_suffixes: List[str] = None, + parse_notes: bool = False, +) -> Dict: + """ + Parse a brat file into the schema defined below. + `txt_file` should be the path to the brat '.txt' file you want to parse, e.g. 'data/1234.txt' + Assumes that the annotations are contained in one or more of the corresponding '.a1', '.a2' or '.ann' files, + e.g. 'data/1234.ann' or 'data/1234.a1' and 'data/1234.a2'. + Will include annotator notes, when `parse_notes == True`. + brat_features = datasets.Features( + { + "id": datasets.Value("string"), + "document_id": datasets.Value("string"), + "text": datasets.Value("string"), + "text_bound_annotations": [ # T line in brat, e.g. type or event trigger + { + "offsets": datasets.Sequence([datasets.Value("int32")]), + "text": datasets.Sequence(datasets.Value("string")), + "type": datasets.Value("string"), + "id": datasets.Value("string"), + } + ], + "events": [ # E line in brat + { + "trigger": datasets.Value( + "string" + ), # refers to the text_bound_annotation of the trigger, + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "arguments": datasets.Sequence( + { + "role": datasets.Value("string"), + "ref_id": datasets.Value("string"), + } + ), + } + ], + "relations": [ # R line in brat + { + "id": datasets.Value("string"), + "head": { + "ref_id": datasets.Value("string"), + "role": datasets.Value("string"), + }, + "tail": { + "ref_id": datasets.Value("string"), + "role": datasets.Value("string"), + }, + "type": datasets.Value("string"), + } + ], + "equivalences": [ # Equiv line in brat + { + "id": datasets.Value("string"), + "ref_ids": datasets.Sequence(datasets.Value("string")), + } + ], + "attributes": [ # M or A lines in brat + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "ref_id": datasets.Value("string"), + "value": datasets.Value("string"), + } + ], + "normalizations": [ # N lines in brat + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "ref_id": datasets.Value("string"), + "resource_name": datasets.Value( + "string" + ), # Name of the resource, e.g. "Wikipedia" + "cuid": datasets.Value( + "string" + ), # ID in the resource, e.g. 534366 + "text": datasets.Value( + "string" + ), # Human readable description/name of the entity, e.g. "Barack Obama" + } + ], + ### OPTIONAL: Only included when `parse_notes == True` + "notes": [ # # lines in brat + { + "id": datasets.Value("string"), + "type": datasets.Value("string"), + "ref_id": datasets.Value("string"), + "text": datasets.Value("string"), + } + ], + }, + ) + """ + + example = {} + example["document_id"] = txt_file.with_suffix("").name + with txt_file.open() as f: + example["text"] = f.read() + + # If no specific suffixes of the to-be-read annotation files are given - take standard suffixes + # for event extraction + if annotation_file_suffixes is None: + annotation_file_suffixes = [".a1", ".a2", ".ann"] + + if len(annotation_file_suffixes) == 0: + raise AssertionError( + "At least one suffix for the to-be-read annotation files should be given!" + ) + + ann_lines = [] + for suffix in annotation_file_suffixes: + annotation_file = txt_file.with_suffix(suffix) + if annotation_file.exists(): + with annotation_file.open() as f: + ann_lines.extend(f.readlines()) + + example["text_bound_annotations"] = [] + example["events"] = [] + example["relations"] = [] + example["equivalences"] = [] + example["attributes"] = [] + example["normalizations"] = [] + + if parse_notes: + example["notes"] = [] + + for line in ann_lines: + line = line.strip() + if not line: + continue + + if line.startswith("T"): # Text bound + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + ann["type"] = fields[1].split()[0] + ann["offsets"] = [] + span_str = remove_prefix(fields[1], (ann["type"] + " ")) + text = fields[2] + for span in span_str.split(";"): + start, end = span.split() + ann["offsets"].append([int(start), int(end)]) + + # Heuristically split text of discontiguous entities into chunks + ann["text"] = [] + if len(ann["offsets"]) > 1: + i = 0 + for start, end in ann["offsets"]: + chunk_len = end - start + ann["text"].append(text[i : chunk_len + i]) + i += chunk_len + while i < len(text) and text[i] == " ": + i += 1 + else: + ann["text"] = [text] + + example["text_bound_annotations"].append(ann) + + elif line.startswith("E"): + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + + ann["type"], ann["trigger"] = fields[1].split()[0].split(":") + + ann["arguments"] = [] + for role_ref_id in fields[1].split()[1:]: + argument = { + "role": (role_ref_id.split(":"))[0], + "ref_id": (role_ref_id.split(":"))[1], + } + ann["arguments"].append(argument) + + example["events"].append(ann) + + elif line.startswith("R"): + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + ann["type"] = fields[1].split()[0] + + ann["head"] = { + "role": fields[1].split()[1].split(":")[0], + "ref_id": fields[1].split()[1].split(":")[1], + } + ann["tail"] = { + "role": fields[1].split()[2].split(":")[0], + "ref_id": fields[1].split()[2].split(":")[1], + } + + example["relations"].append(ann) + + # '*' seems to be the legacy way to mark equivalences, + # but I couldn't find any info on the current way + # this might have to be adapted dependent on the brat version + # of the annotation + elif line.startswith("*"): + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + ann["ref_ids"] = fields[1].split()[1:] + + example["equivalences"].append(ann) + + elif line.startswith("A") or line.startswith("M"): + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + + info = fields[1].split() + ann["type"] = info[0] + ann["ref_id"] = info[1] + + if len(info) > 2: + ann["value"] = info[2] + else: + ann["value"] = "" + + example["attributes"].append(ann) + + elif line.startswith("N"): + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + ann["text"] = fields[2] + + info = fields[1].split() + + ann["type"] = info[0] + ann["ref_id"] = info[1] + ann["resource_name"] = info[2].split(":")[0] + ann["cuid"] = info[2].split(":")[1] + example["normalizations"].append(ann) + + elif parse_notes and line.startswith("#"): + ann = {} + fields = line.split("\t") + + ann["id"] = fields[0] + ann["text"] = fields[2] if len(fields) == 3 else BigBioValues.NULL + + info = fields[1].split() + + ann["type"] = info[0] + ann["ref_id"] = info[1] + example["notes"].append(ann) + + return example + + +def brat_parse_to_bigbio_kb(brat_parse: Dict) -> Dict: + """ + Transform a brat parse (conforming to the standard brat schema) obtained with + `parse_brat_file` into a dictionary conforming to the `bigbio-kb` schema (as defined in ../schemas/kb.py) + :param brat_parse: + """ + + unified_example = {} + + # Prefix all ids with document id to ensure global uniqueness, + # because brat ids are only unique within their document + id_prefix = brat_parse["document_id"] + "_" + + # identical + unified_example["document_id"] = brat_parse["document_id"] + unified_example["passages"] = [ + { + "id": id_prefix + "_text", + "type": "abstract", + "text": [brat_parse["text"]], + "offsets": [[0, len(brat_parse["text"])]], + } + ] + + # get normalizations + ref_id_to_normalizations = defaultdict(list) + for normalization in brat_parse["normalizations"]: + ref_id_to_normalizations[normalization["ref_id"]].append( + { + "db_name": normalization["resource_name"], + "db_id": normalization["cuid"], + } + ) + + # separate entities and event triggers + unified_example["events"] = [] + non_event_ann = brat_parse["text_bound_annotations"].copy() + for event in brat_parse["events"]: + event = event.copy() + event["id"] = id_prefix + event["id"] + trigger = next( + tr + for tr in brat_parse["text_bound_annotations"] + if tr["id"] == event["trigger"] + ) + if trigger in non_event_ann: + non_event_ann.remove(trigger) + event["trigger"] = { + "text": trigger["text"].copy(), + "offsets": trigger["offsets"].copy(), + } + for argument in event["arguments"]: + argument["ref_id"] = id_prefix + argument["ref_id"] + + unified_example["events"].append(event) + + unified_example["entities"] = [] + anno_ids = [ref_id["id"] for ref_id in non_event_ann] + for ann in non_event_ann: + entity_ann = ann.copy() + entity_ann["id"] = id_prefix + entity_ann["id"] + entity_ann["normalized"] = ref_id_to_normalizations[ann["id"]] + unified_example["entities"].append(entity_ann) + + # massage relations + unified_example["relations"] = [] + skipped_relations = set() + for ann in brat_parse["relations"]: + if ( + ann["head"]["ref_id"] not in anno_ids + or ann["tail"]["ref_id"] not in anno_ids + ): + skipped_relations.add(ann["id"]) + continue + unified_example["relations"].append( + { + "arg1_id": id_prefix + ann["head"]["ref_id"], + "arg2_id": id_prefix + ann["tail"]["ref_id"], + "id": id_prefix + ann["id"], + "type": ann["type"], + "normalized": [], + } + ) + if len(skipped_relations) > 0: + example_id = brat_parse["document_id"] + logger.info( + f"Example:{example_id}: The `bigbio_kb` schema allows `relations` only between entities." + f" Skip (for now): " + f"{list(skipped_relations)}" + ) + + # get coreferences + unified_example["coreferences"] = [] + for i, ann in enumerate(brat_parse["equivalences"], start=1): + is_entity_cluster = True + for ref_id in ann["ref_ids"]: + if not ref_id.startswith("T"): # not textbound -> no entity + is_entity_cluster = False + elif ref_id not in anno_ids: # event trigger -> no entity + is_entity_cluster = False + if is_entity_cluster: + entity_ids = [id_prefix + i for i in ann["ref_ids"]] + unified_example["coreferences"].append( + {"id": id_prefix + str(i), "entity_ids": entity_ids} + ) + return unified_example diff --git a/bigbio/hub/hub_repos/s800/s800.py b/bigbio/hub/hub_repos/s800/s800.py new file mode 100644 index 00000000..8e150daf --- /dev/null +++ b/bigbio/hub/hub_repos/s800/s800.py @@ -0,0 +1,253 @@ +# coding=utf-8 +# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +""" +S800 Corpus: a novel abstract-based manually annotated corpus for Named Entity Recognition. +S800 comprises 800 PubMed abstracts in which organism mentions were identified and mapped +to the corresponding NCBI Taxonomy identifiers. + +To increase the corpus taxonomic mention diversity the S800 abstracts were collected by +selecting 100 abstracts from the following 8 categories: +bacteriology, botany, entomology, medicine, mycology, protistology, virology and zoology. +S800 has been annotated with a focus at the species level; +however, higher taxa mentions (such as genera, families and orders) have also been considered. +""" + +from pathlib import Path +from typing import Any, Dict, List, Tuple + +import datasets +import pandas as pd + +from .bigbiohub import BigBioConfig, Tasks, kb_features + +_LOCAL = False + +_CITATION = """\ +@article{ + title = {The SPECIES and ORGANISMS Resources for Fast and Accurate Identification of Taxonomic Names in Text}, + author = {Pafilis, Evangelos AND Frankild, Sune P. AND Fanini, + Lucia AND Faulwetter, Sarah AND Pavloudi, Christina AND Vasileiadou, + Aikaterini AND Arvanitidis, Christos AND Jensen, Lars Juhl}, + journal = {PLOS ONE}, + publisher = {Public Library of Science}, + year = {2013}, + month = {06}, + volume = {8}, + pages = {1-6}, + number = {6}, + url = {https://doi.org/10.1371/journal.pone.0065390}, + doi = {10.1371/journal.pone.0065390}, +} +""" + +_DATASETNAME = "s800" + +_DISPLAYNAME = "S800" + +_DESCRIPTION = """\ +S800 Corpus: a novel abstract-based manually annotated corpus for Named Entity Recognition. +S800 comprises 800 PubMed abstracts in which organism mentions were identified and mapped +to the corresponding NCBI Taxonomy identifiers. + +To increase the corpus taxonomic mention diversity the S800 abstracts were collected by +selecting 100 abstracts from the following 8 categories: +bacteriology, botany, entomology, medicine, mycology, protistology, virology and zoology. +S800 has been annotated with a focus at the species level; +however, higher taxa mentions (such as genera, families and orders) have also been considered. +""" + +_HOMEPAGE = "https://species.jensenlab.org/" + +_LICENSE = "OTHER" # "subject to Medline restrictions" + +_URLS = { + _DATASETNAME: "https://species.jensenlab.org/files/S800-1.0.tar.gz", +} + +_LANGUAGES = ["English"] + +_PUBMED = True + +_SUPPORTED_TASKS = [Tasks.NAMED_ENTITY_RECOGNITION, Tasks.NAMED_ENTITY_DISAMBIGUATION] + +_SOURCE_VERSION = "1.0.0" + +_BIGBIO_VERSION = "1.0.0" + + +class S800Dataset(datasets.GeneratorBasedBuilder): + """S800 comprises 800 PubMed abstracts in which organism mentions + were identified and mapped to the corresponding NCBI Taxonomy identifiers.""" + + SOURCE_VERSION = datasets.Version(_SOURCE_VERSION) + BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION) + + BUILDER_CONFIGS = [ + BigBioConfig( + name=f"{_DATASETNAME}_source", + version=SOURCE_VERSION, + description=f"{_DATASETNAME} source schema", + schema="source", + subset_id=f"{_DATASETNAME}", + ), + BigBioConfig( + name=f"{_DATASETNAME}_bigbio_kb", + version=BIGBIO_VERSION, + description=f"{_DATASETNAME} BigBio schema", + schema="bigbio_kb", + subset_id=f"{_DATASETNAME}", + ), + ] + + DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source" + + def _info(self) -> datasets.DatasetInfo: + if self.config.schema == "source": + features = datasets.Features( + { + "doc_id": datasets.Value("string"), + "s800_doc_id": datasets.Value("string"), + "pmid": datasets.Value("string"), + "entities": [ + { + "offsets": datasets.Sequence(datasets.Value("int64")), + "text": datasets.Value("string"), + "ncbi_txid": datasets.Value("string"), + } + ], + "category": datasets.Value("string"), + "category_id": datasets.Value("int64"), + "journal": datasets.Value("string"), + "text": datasets.Value("string"), + } + ) + + elif self.config.schema == "bigbio_kb": + features = kb_features + + return datasets.DatasetInfo( + description=_DESCRIPTION, + features=features, + homepage=_HOMEPAGE, + license=_LICENSE, + citation=_CITATION, + ) + + def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]: + """Returns SplitGenerators.""" + urls = _URLS[_DATASETNAME] + data_dir = dl_manager.download_and_extract(urls) + + return [ + datasets.SplitGenerator( + name=datasets.Split.TRAIN, + # Whatever you put in gen_kwargs will be passed to _generate_examples + gen_kwargs={ + "data_dir": Path(data_dir), + "split": "train", + }, + ), + ] + + def _generate_examples(self, data_dir: Path, split: str) -> Tuple[int, Dict]: + """Yields examples as (key, example) tuples.""" + if self.config.schema == "source": + for key, example in self._read_example_from_file(data_dir): + yield key, example + + elif self.config.schema == "bigbio_kb": + for key, example in self._read_example_from_file_in_kb_schema(data_dir): + yield key, example + + def _read_example_from_file(self, data_dir: Path) -> Tuple[str, Dict]: + abstract_dir = data_dir / "abstracts" + df_s800 = pd.read_csv( + data_dir / "S800.tsv", + sep="\t", + header=None, + names=["nbci_taxonomy_id", "doc_id", "start", "end", "phrase"], + ) + df_s800["s800_doc_id"] = df_s800["doc_id"].apply(lambda x: x.split(":")[0]) + + df_pubmed = pd.read_csv( + data_dir / "pubmedid.tsv", + sep="\t", + header=None, + names=["s800_doc_id", "pmid", "category", "category_id", "journal"], + ) + for _, row in df_pubmed.iterrows(): + key = row.s800_doc_id + entities = [ + dict( + offsets=[entity_row.start, entity_row.end], + text=entity_row.phrase, + ncbi_txid=entity_row.nbci_taxonomy_id, + ) + for _, entity_row in df_s800[df_s800.s800_doc_id == key].iterrows() + ] + doc_abstract_path = abstract_dir / f"{key}.txt" + with open(doc_abstract_path, encoding="utf-8") as fp: + text = fp.read() + example = { + "doc_id": key, + "s800_doc_id": key, + "pmid": row.pmid, + "entities": entities, + "category": row.category, + "category_id": row.category_id, + "journal": row.journal, + "text": text, + } + yield key, example + + def _parse_example_to_kb_schema(self, example) -> Dict[str, Any]: + text = example["text"] + doc_id = example["doc_id"] + passages = [ + { + "id": f"{doc_id}-P0", + "type": "abstract", + "text": [text], + "offsets": [[0, len(text)]], + } + ] + entities = [] + for i, entity in enumerate(example["entities"]): + cs, ce = entity["offsets"] + ce = ce + 1 # Add 1 to make the offset exclusive + entity = { + "id": f"{doc_id}-E{i}", + "text": [entity["text"]], + "offsets": [[cs, ce]], + "type": "species", + "normalized": [{"db_id": entity["ncbi_txid"], "db_name": "NBCI Taxonomy"}], + } + entities.append(entity) + data = { + "id": doc_id, + "document_id": doc_id, + "passages": passages, + "entities": entities, + "relations": [], + "events": [], + "coreferences": [], + } + return data + + def _read_example_from_file_in_kb_schema(self, data_dir: Path) -> Tuple[str, Dict]: + for key, example in self._read_example_from_file(data_dir): + example = self._parse_example_to_kb_schema(example) + yield key, example diff --git a/biodatasets/s800/s800.py b/biodatasets/s800/s800.py new file mode 100644 index 00000000..73f240ed --- /dev/null +++ b/biodatasets/s800/s800.py @@ -0,0 +1,254 @@ +# coding=utf-8 +# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +""" +S800 Corpus: a novel abstract-based manually annotated corpus for Named Entity Recognition. +S800 comprises 800 PubMed abstracts in which organism mentions were identified and mapped to the corresponding NCBI Taxonomy identifiers. + +To increase the corpus taxonomic mention diversity the S800 abstracts were collected by selecting 100 abstracts from the following 8 categories: bacteriology, botany, entomology, medicine, mycology, protistology, virology and zoology. +S800 has been annotated with a focus at the species level; however, higher taxa mentions (such as genera, families and orders) have also been considered. +""" + +import os +from pathlib import Path +from typing import Any, List, Tuple, Dict + +import datasets +import pandas as pd +from bigbio.utils import schemas +from bigbio.utils.configs import BigBioConfig +from bigbio.utils.constants import Tasks + +_CITATION = """\ +@article{, + title = {The SPECIES and ORGANISMS Resources for Fast and Accurate Identification of Taxonomic Names in Text}, + author = {Pafilis, Evangelos AND Frankild, Sune P. AND Fanini, Lucia AND Faulwetter, Sarah AND Pavloudi, Christina AND Vasileiadou, Aikaterini AND Arvanitidis, Christos AND Jensen, Lars Juhl}, + journal = {PLOS ONE}, + publisher = {Public Library of Science}, + year = {2013}, + month = {06}, + volume = {8}, + pages = {1-6}, + number = {6}, + url = {https://doi.org/10.1371/journal.pone.0065390}, + doi = {10.1371/journal.pone.0065390}, + biburl = {https://journals.plos.org/plosone/article/citation/bibtex?id=10.1371/journal.pone.0065390}, + bibsource = {https://journals.plos.org/plosone/article/citation?id=10.1371/journal.pone.0065390} +} +""" + +_DATASETNAME = "s800" + +_DESCRIPTION = """\ +S800 Corpus: a novel abstract-based manually annotated corpus. +S800 comprises 800 PubMed abstracts in which organism mentions were identified and mapped to the corresponding NCBI Taxonomy identifiers. + +To increase the corpus taxonomic mention diversity the S800 abstracts were collected by selecting 100 abstracts from the following 8 categories: bacteriology, botany, entomology, medicine, mycology, protistology, virology and zoology. +S800 has been annotated with a focus at the species level; however, higher taxa mentions (such as genera, families and orders) have also been considered. +""" + +_HOMEPAGE = "https://species.jensenlab.org/" + +_LICENSE = "Creative Commons License Attribution-ShareAlike 4.0 International" + +_URLS = { + _DATASETNAME: "https://species.jensenlab.org/files/S800-1.0.tar.gz", +} + +_SUPPORTED_TASKS = [Tasks.NAMED_ENTITY_RECOGNITION] + +_SOURCE_VERSION = "1.0.0" + +_BIGBIO_VERSION = "1.0.0" + + +class S800Dataset(datasets.GeneratorBasedBuilder): + """S800 comprises 800 PubMed abstracts in which organism mentions were identified and mapped to the corresponding NCBI Taxonomy identifiers.""" + + SOURCE_VERSION = datasets.Version(_SOURCE_VERSION) + BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION) + + BUILDER_CONFIGS = [ + BigBioConfig( + name=f"{_DATASETNAME}_source", + version=SOURCE_VERSION, + description=f"{_DATASETNAME} source schema", + schema="source", + subset_id=f"{_DATASETNAME}", + ), + BigBioConfig( + name=f"{_DATASETNAME}_bigbio_kb", + version=BIGBIO_VERSION, + description=f"{_DATASETNAME} BigBio schema", + schema="bigbio_kb", + subset_id=f"{_DATASETNAME}", + ), + ] + + DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source" + + def _info(self) -> datasets.DatasetInfo: + if self.config.schema == "source": + features = datasets.Features( + { + "doc_id": datasets.Value("string"), + "s800_doc_id": datasets.Value("string"), + "pmid": datasets.Value("string"), + "entities": { + "offsets": [datasets.Value("int64")], + "text": datasets.Value("string"), + "ncbi_txid": datasets.Value("string"), + }, + "category": datasets.Value("string"), + "category_id": datasets.Value("int64"), + "journal": datasets.Value("string"), + "text": datasets.Value("string"), + } + ) + + elif self.config.schema == "bigbio_kb": + features = schemas.kb_features + + return datasets.DatasetInfo( + description=_DESCRIPTION, + features=features, + homepage=_HOMEPAGE, + license=_LICENSE, + citation=_CITATION, + ) + + def _split_generators(self, dl_manager) -> List[datasets.SplitGenerator]: + """Returns SplitGenerators.""" + urls = _URLS[_DATASETNAME] + data_dir = dl_manager.download_and_extract(urls) + + return [ + datasets.SplitGenerator( + name=datasets.Split.TRAIN, + # Whatever you put in gen_kwargs will be passed to _generate_examples + gen_kwargs={ + "data_dir": Path(data_dir), + "split": "train", + }, + ), + ] + + def _generate_examples(self, data_dir: Path, split: str) -> Tuple[int, Dict]: + """Yields examples as (key, example) tuples.""" + if self.config.schema == "source": + for key, example in self._read_example_from_file(data_dir): + yield key, example + + elif self.config.schema == "bigbio_kb": + for key, example in self._read_example_from_file_in_kb_schema(data_dir): + yield key, example + + def _read_example_from_file(self, data_dir: Path) -> Tuple[str, Dict]: + abstract_dir = data_dir / "abstracts" + df_s800 = pd.read_csv( + data_dir / "S800.tsv", + sep="\t", + header=None, + names=["nbci_taxonomy_id", "doc_id", "start", "end", "phrase"], + ).assign( + ncbi_txid=lambda dft: dft["nbci_taxonomy_id"].apply( + lambda x: f"NCBI:txid{x}" + ) + ) + + df_pubmed = pd.read_csv( + data_dir / "pubmedid.tsv", + sep="\t", + header=None, + names=["s800_doc_id", "pmid", "category", "category_id", "journal"], + ) + + df = ( + df_s800.groupby("doc_id") + .agg(list) + .reset_index() + .merge( + df_pubmed.assign( + doc_id=lambda dft: ( + dft["s800_doc_id"] + ":" + dft["pmid"] + ).str.replace("PMID:", "") + ), + on="doc_id", + how="left", + ) + ) + for _, row in df.iterrows(): + key = row.doc_id + entities = [ + dict(offsets=[s, e], text=p, ncbi_txid=ncbi_txid) + for s, e, p, ncbi_txid in zip( + row.start, row.end, row.phrase, row.ncbi_txid + ) + ] + doc_abstract_path = abstract_dir / f"{row.s800_doc_id}.txt" + with open(doc_abstract_path, encoding="utf-8") as fp: + text = fp.read() + example = { + "doc_id": key, + "s800_doc_id": row.s800_doc_id, + "pmid": row.pmid, + "entities": entities, + "category": row.category, + "category_id": row.category_id, + "journal": row.journal, + "text": text, + } + yield key, example + + def _parse_example_to_kb_schema(self, example) -> Dict[str, Any]: + text = example["text"] + doc_id = example["doc_id"] + passages = [ + { + "id": f"{doc_id}-P0", + "type": "abstract", + "text": [text], + "offsets": [[0, len(text)]], + } + ] + entities = [] + for i, entity in enumerate(example["entities"]): + cs, ce = entity["offsets"] + ce = ce + 1 # Add 1 to make the offset exclusive + entity = { + "id": f"{doc_id}-E{i}", + "text": [entity["text"]], + "offsets": [[cs, ce]], + "type": "species", + "normalized": [ + {"db_id": entity["ncbi_txid"], "db_name": "NBCI Taxonomy"} + ], + } + entities.append(entity) + data = { + "id": doc_id, + "document_id": doc_id, + "passages": passages, + "entities": entities, + "relations": [], + "events": [], + "coreferences": [], + } + return data + + def _read_example_from_file_in_kb_schema(self, data_dir: Path) -> Tuple[str, Dict]: + for key, example in self._read_example_from_file(data_dir): + example = self._parse_example_to_kb_schema(example) + yield key, example