scripts/adjust_annotation_end.py

# Author: Roman Janík
# Script for adjusting the end of entity annotation span in text. Label Studio marks additional characters
# at the end of entity, which do not belong to it (usually dot or comma).
#

import argparse
import json
import re

from pynput import keyboard
from pynput.keyboard import Key


def get_args():
    parser = argparse.ArgumentParser(description="Script for adjusting the end of entity annotation span in text. Label\
                                      Studio marks additional characters at the end of entity, which do not belong to it\
                                      (usually dot or comma).")
    parser.add_argument("-s", "--source_file", required=True, help="Path to source Label Studio json annotations file.")
    args = parser.parse_args()
    return args


def save(annotations_file, annotations):
    with open(annotations_file, "w", encoding='utf8') as f:
        json.dump(annotations, f, indent=2, ensure_ascii=False)
    print("Annotations were saved!")


def auto_correct(n_entity_text, n_entity_type):
    if len(n_entity_text) >= 2 and n_entity_text[-2].isalnum() and n_entity_text[-1] == "\n":
        return 1

    if n_entity_type == "Artifact names/Objects":
        monetary_u = ["zl.", "kr.", "zl. r. m.", "kr. r. m.", "tol.", "zl. r. č.", "h.", "hal.", "fl.", "gr."]
        if n_entity_text in monetary_u:
            return 0
        to_be_determined = ["Uh.", "př.", "č.", "R.", "lt.", "r.", "J. K. K.", "sv. Karla\nBor.",
                            "čís. 80. Sov. zák. a nař.", "čís. 80\nSb. z. a n."]
        if n_entity_text in to_be_determined:
            return -1
        if n_entity_text.endswith("Sb."):
            return 0
        if len(n_entity_text) >= 2 and n_entity_text[-2].isalnum():
            return 1

    if n_entity_type == "Time expressions":
        # year
        if re.search("^[0-9]{4}$", n_entity_text[:-1]):
            return 1
        # single day
        if re.search("^[0-9][.]$|^[0-9]{2}[.]$", n_entity_text):
            return 0
        # single day with comma
        if re.search("^[0-9][,]$|^[0-9]{2}[,]$", n_entity_text):
            return 1
        # day and month
        if re.search("^[0-9][.][ ]?[0-9][.]$|^[0-9][.][ ]?[0-9]{2}[.]$|^[0-9]{2}[.][ ]?[0-9][.]$|^[0-9]{2}[.][ ]?["
                     "0-9]{2}[.]$", n_entity_text):
            return 0
        # year span
        if re.search("^[0-9]{4}[ ]?[-][ ]?[0-9]{4}$|^[0-9]{4}[ ]?[-][ ]?[0-9]{2}$", n_entity_text[:-1]):
            return 1
        # date with "hod." or "t.r." at the end
        if n_entity_text.endswith(("hod.", "t.r.", "t. r.", "t.m.", "t. m.")):
            return 0

    if n_entity_type == "Geographical names":
        # street name correct
        if n_entity_text.endswith(("ul.", "tř.", "nám.")):
            return 0
        # street name incorrect
        if n_entity_text.endswith(("ul.,", "tř.,", "nám.,")):
            return 1

    if n_entity_type == "Personal names":
        # names ending with "ml." or "st." correct
        if n_entity_text.endswith(("ml.", "st.")):
            return 0
        # names ending with "ml." or "st." incorrect
        if n_entity_text.endswith(("ml.,", "st.,")):
            return 1

    if n_entity_type == "Institutions":
        # M.N.V., O.N.V., K.N.V., N.J., N.F., J.Z.D.
        if re.search("^[M][.][ ]?[N][.][ ]?[V][.]$|^[O][.][ ]?[N][.][ ]?[V][.]$|^[K][.][ ]?[N][.][ ]?[V][.]$|^[N][.][ "
                     "]?[J][.]$|^[N][.][ ]?[F][.]|^[J][.][ ]?[Z][.][ ]?[D][.]$", n_entity_text):
            return 0
        # M.N.V., O.N.V., K.N.V., N.J., N.F., J.Z.D. with additional char
        if re.search(
                "^[M][.][ ]?[N][.][ ]?[V][.]$|^[O][.][ ]?[N][.][ ]?[V][.]$|^[K][.][ ]?[N][.][ ]?[V][.]$|^[N][.][ "
                "]?[J][.]$|^[N][.][ ]?[F][.]|^[J][.][ ]?[Z][.][ ]?[D][.]$", n_entity_text[:-1]):
            return 1
        if n_entity_text[:-1] in ["JZD", "KSČ", "Ksč", "ksč", "Kčs", "ksč", "NF", "MNV", "ONV", "KNV"]:
            return 1

    if n_entity_type in ["Geographical names", "Personal names", "Institutions"]:
        # shorten entities with non-dot char at the end
        if n_entity_text[-1] != ".":
            words = n_entity_text[:-1].split()
            if all([word.isalnum() for word in words]):
                return 1

    return -1


def main():
    args = get_args()

    print("Script for adjusting the end of entity annotation span in text. "
          "Script goes through page text files and their annotations json record. "
          "End of entity is adjusted manually or automatically is possible and annotations are saved to the same file."
          "Adjusting starts at last adjusted entity, controls: right arrow - next entity, left arrow - previous entity,"
          "up arrow - +1 length, down arrow - -1 length, s - save"
          "automatic save after 10 entities\n")

    def on_key_release(key):
        nonlocal annotations, n_entity_idx, page_idx, page_text, stop_edit, adjusted_n_entities

        # next entity
        if key == Key.right:
            annotations[page_idx]["ner"][n_entity_idx]["adjusted"] = True
            n_entity_idx += 1
            adjusted_n_entities += 1
            exit()
        # previous adjusted entity
        elif key == Key.left:
            while n_entity_idx > 0:
                n_entity_idx -= 1
                if "adjusted" in annotations[page_idx]["ner"][n_entity_idx].keys():
                    annotations[page_idx]["ner"][n_entity_idx]["adjusted"] = False
                    adjusted_n_entities -= 1
                    break
            exit()
        # +1 length
        elif key == Key.up:
            if annotations[page_idx]["ner"][n_entity_idx]["end"] + 1 == len(page_text):
                print("Entity span cannot be prolonged, end of page text is reached!")
            else:
                annotations[page_idx]["ner"][n_entity_idx]["end"] += 1
                print("Entity length + 1, end:   {}".format(annotations[page_idx]["ner"][n_entity_idx]["end"]))
        # -1 length
        elif key == Key.down:
            if annotations[page_idx]["ner"][n_entity_idx]["end"] - 1 == 0:
                print("Entity span cannot be shortened, start of page text is reached!")
            else:
                annotations[page_idx]["ner"][n_entity_idx]["end"] -= 1
                print("Entity length - 1, end:   {}".format(annotations[page_idx]["ner"][n_entity_idx]["end"]))
        # save
        elif key == Key.esc:
            save(args.source_file, annotations)
            print("Editing stopped!")
            stop_edit = True
            exit()

    with open(args.source_file, encoding="utf-8") as f:
        annotations = json.load(f)

    page_idx = 0
    n_entity_idx = 0
    for i, page in enumerate(annotations):
        if "adjusted" not in page.keys() or not page["adjusted"]:
            page_idx = i
            for j, n_entity in enumerate(page["ner"]):
                if "adjusted" not in n_entity.keys():
                    n_entity_idx = j
                    break
            break

    adjusted_pages = 0
    adjusted_n_entities = 1
    stop_edit = False
    while True:
        page_text_path = annotations[page_idx]["text"].replace(
            "http://localhost:8081", "../../../datasets/poner1.0/data")
        with open(page_text_path, encoding="utf-8") as p_f:
            page_text = p_f.read()
        page_name = annotations[page_idx]["page_name"]
        print(f"Page:   {adjusted_pages}\n{page_name}\n{page_text}\n\nEntities without alphanum end:\n")

        while True:
            n_entity = annotations[page_idx]["ner"][n_entity_idx]
            n_entity_text = page_text[n_entity["start"]:n_entity["end"]]
            n_entity_type = n_entity["labels"][0]
            if not n_entity_text[-1].isalnum() or ("adjusted" in n_entity.keys() and not n_entity["adjusted"]):
                # try auto correct function, if auto correction is not possible, manual correction is applied
                end_shift = auto_correct(n_entity_text, n_entity_type)
                if end_shift == -1 or ("adjusted" in n_entity.keys() and not n_entity["adjusted"]):
                    context_start = 0 if n_entity["start"] - 100 < 0 else n_entity["start"] - 100
                    context_end = len(page_text) - 1 if n_entity["start"] + 100 >= len(page_text) \
                        else n_entity["start"] + 100
                    print(
                        f"\n{n_entity_text}\n------------------------------\n{n_entity_type}\n"
                        f"------------------------------\n{page_text[context_start:context_end]}\n")
                    with keyboard.Listener(on_release=on_key_release) as listener:
                        listener.join()
                    if stop_edit:
                        return
                else:
                    # auto correction application
                    annotations[page_idx]["ner"][n_entity_idx]["end"] -= end_shift
                    annotations[page_idx]["ner"][n_entity_idx]["auto_adjusted"] = True
                    n_entity_idx += 1
                    adjusted_n_entities += 1

                # auto save
                if adjusted_n_entities % 10 == 0:
                    save(args.source_file, annotations)
            else:
                n_entity_idx += 1
            if n_entity_idx == len(annotations[page_idx]["ner"]):
                n_entity_idx = 0
                break

        adjusted_pages += 1

        annotations[page_idx]["adjusted"] = True
        page_idx += 1
        if page_idx == len(annotations):
            save(args.source_file, annotations)
            break

    print("All pages were adjusted!")


if __name__ == '__main__':
    main()