create_dataset.py

import os
import json
import glob
from tqdm import tqdm
from io import StringIO
import pandas as pd

"""
Steps to convert the data into JSON format.

Step-0: Use a python environment where pandas is installed.
Step-1: Download the source file from here: https://figshare.com/articles/dataset/MACCROBAT2018/9764942
Step-2: Unzip the file in put that into a folder (say `data` folder).
    All unzipped files will be present here.
        * data/MACCROBAT2020/*.txt
        * data/MACCROBAT2020/*.ann
Step-3: Use the correct paths and run this file.
"""

def remove_overlapped_ner_tags(ner_details: list[dict]):
    """remove overlapping entities.

    Args:
        ner_details (List[dict]): a list of dictionary where each dictionary holds
            the information of a entity.

    NOTE: Priority is given to the entity that is labelled first after sorting all by start index in ascending order.
    (i.e. it's end-index is less than other start of other overlapping entity.)

    Returns:
        list[dict]: updated list (removed item if something was overlapping)
    """
    # funtion to remove the overlapping NER-tags
    new_ner_details = []

    ner_details = sorted(ner_details, key=lambda x: x["start"])
    for i, ner_detail in enumerate(ner_details):
        if i == 0:
            start = ner_detail["start"]
            end = ner_detail["end"]
            new_ner_details.append(ner_detail)
            continue

        current_start = ner_detail["start"]
        current_end = ner_detail["end"]
        if current_start < end:
            continue

        # update the start and end
        start = current_start
        end = current_end

        new_ner_details.append(ner_detail)
    return new_ner_details


def get_ner_details(ann_file):
    with open(ann_file, "r") as f:
        lines = f.readlines()
        lines = [line.strip() for line in lines]
        csv_data = "\n".join(lines)
        csv_data = StringIO(csv_data)
    df = pd.read_csv(csv_data, sep="\t", header=None)

    df.columns = ["EntityID", "EntityDetails", "EntityText"]
    # print(df.shape)

    # remove rows where entity-id start other than `T`
    df = df[df["EntityID"].apply(lambda x: str(x).strip().startswith("T"))]

    # remove the rows which contains the ";" in the `EntityDetails`
    df = df[df["EntityDetails"].apply(lambda x: ";" not in str(x))]

    # drop where None is present
    df.dropna(axis=1, inplace=True)

    ner_info = []
    for i, row in df.iterrows():
        text = row["EntityText"]
        details = row["EntityDetails"]
        try:
            ner_tag, start, end = details.split(" ")
        except:
            print(ann_file)
            print(details)
        start = int(float(start))
        end = int(float(end))

        ner_info.append({"text": text, "label": ner_tag.upper(), "start": start, "end": end})

    # remove the overlapping entities
    ner_info = remove_overlapped_ner_tags(ner_details=ner_info)
    # print(ner_info)

    return ner_info


def main(input_path: str = "data/MACCROBAT2020", output_path: str = "data/MACCROBAT2020-V2.json"):
    txt_files = glob.glob(os.path.join(input_path, "*.txt"))
    txt_files.sort()

    ner_data = {"data": [], "verson": "MACCROBAT-V2 (https://figshare.com/articles/dataset/MACCROBAT2018/9764942)"}

    for txt_file in tqdm(txt_files, desc="Extracting data..."):
        with open(txt_file, "r") as f:
            full_text = f.read()
        a = txt_file.replace(".txt", ".ann")
        ner_info = get_ner_details(a)
        data = {"full_text": full_text, "ner_info": ner_info}
        ner_data["data"].append(data)

    ALL_NER_LABLES = set()
    for details in tqdm(ner_data["data"], desc="Splitting into tokens..."):
        text = details["full_text"]
        ner_details = details["ner_info"]

        tokens = []
        ner_labels = []
        start = 0
        for ner_detail in ner_details:
            ner_start = ner_detail["start"]
            ner_end = ner_detail["end"]

            before_ner_token = text[start:ner_start]
            ner_token = text[ner_start:ner_end]

            tokens.append(before_ner_token)
            ner_labels.append("O")
            tokens.append(ner_token)
            ner_labels.append(f'B-{ner_detail["label"]}')
            ALL_NER_LABLES.add(f'B-{ner_detail["label"]}')
            ALL_NER_LABLES.add(f'I-{ner_detail["label"]}')
            start = ner_end
        if len(text) >= start:
            ner_labels.append("O")
            tokens.append(text[start:])
        assert len(tokens) == len(ner_labels)

        details["tokens"] = tokens
        details["ner_labels"] = ner_labels

    ner_data["all_ner_labels"] = sorted(list(ALL_NER_LABLES), key=lambda x: x.split("-")[-1])

    label_2_index = {k: i for i, k in enumerate(ner_data["all_ner_labels"])}
    index_2_label = {v: k for k, v in label_2_index.items()}
    ner_data["label_2_index"] = label_2_index
    ner_data["index_2_label"] = index_2_label

    for details in tqdm(ner_data["data"], desc="label2index..."):
        ner_labels = details["ner_labels"]
        ner_labels_ids = []
        for ner in ner_labels:
            ner_labels_ids.append(label_2_index.get(ner))
        details["ner_labels"] = ner_labels_ids
    with open(output_path, "w") as f:
        json.dump(ner_data, f, indent=4)


if __name__ == "__main__":
    input_path: str = "data/MACCROBAT2020"
    output_path: str = "data/MACCROBAT2020-V2.json"
    main(input_path=input_path, output_path=output_path)