|
|
|
|
|
|
|
|
|
import argparse |
|
import json |
|
import os |
|
|
|
from nltk.tokenize import word_tokenize, sent_tokenize |
|
|
|
|
|
def get_args(): |
|
parser = argparse.ArgumentParser(description="Script for creating a CoNLL format of my dataset from text files and Label Studio annotations.") |
|
parser.add_argument("-s", "--source_file", required=True, help="Path to source Label Studio json annotations file.") |
|
parser.add_argument("-o", "--output_dir", required=True, help="Output dir for CoNLL dataset splits.") |
|
args = parser.parse_args() |
|
return args |
|
|
|
|
|
def fix_quotes(sentence): |
|
return map(lambda word: word.replace('``', '"').replace("''", '"'), sentence) |
|
|
|
|
|
def process_text_annotations(text, annotations): |
|
entity_types_map = { |
|
"Personal names": "p", |
|
"Institutions": "i", |
|
"Geographical names": "g", |
|
"Time expressions": "t", |
|
"Artifact names/Objects": "o" |
|
} |
|
|
|
sentences = sent_tokenize(text, language="czech") |
|
sentences_t = [fix_quotes(word_tokenize(x, language="czech")) for x in sentences] |
|
|
|
sentences_idx = [] |
|
start = 0 |
|
for i, c in enumerate(text): |
|
if not c.isspace(): |
|
start = i |
|
break |
|
|
|
for sentence in sentences_t: |
|
sentence_idx = [] |
|
for word in sentence: |
|
end = start + len(word) |
|
sentence_idx.append({"word": word, "start": start, "end": end, "entity_type": "O"}) |
|
for i, _ in enumerate(text): |
|
if end + i < len(text) and not text[end + i].isspace(): |
|
start = end + i |
|
break |
|
sentences_idx.append(sentence_idx) |
|
|
|
for n_entity in annotations: |
|
begin = True |
|
done = False |
|
for sentence_idx in sentences_idx: |
|
for word_idx in sentence_idx: |
|
if word_idx["start"] >= n_entity["start"]\ |
|
and (word_idx["end"] <= n_entity["end"] |
|
or (not text[word_idx["end"]-1].isalnum()) and len(word_idx["word"]) > 1) and begin: |
|
word_idx["entity_type"] = "B-" + entity_types_map[n_entity["labels"][0]] |
|
begin = False |
|
if word_idx["end"] >= n_entity["end"]: |
|
done = True |
|
break |
|
elif word_idx["start"] > n_entity["start"] and (word_idx["end"] <= n_entity["end"] |
|
or (not text[word_idx["end"]-1].isalnum() and text[word_idx["start"]].isalnum())): |
|
word_idx["entity_type"] = "I-" + entity_types_map[n_entity["labels"][0]] |
|
if word_idx["end"] >= n_entity["end"]: |
|
done = True |
|
break |
|
elif word_idx["end"] > n_entity["end"]: |
|
done = True |
|
break |
|
if done: |
|
break |
|
|
|
conll_sentences = [] |
|
for sentence_idx in sentences_idx: |
|
conll_sentence = map(lambda w: w["word"] + " " + w["entity_type"], sentence_idx) |
|
conll_sentences.append("\n".join(conll_sentence)) |
|
conll_sentences = "\n\n".join(conll_sentences) |
|
|
|
return conll_sentences |
|
|
|
|
|
def main(): |
|
args = get_args() |
|
|
|
print("Script for creating a CoNLL format of my dataset from text files and Label Studio annotations." |
|
"Script goes through page text files and their annotations json record. " |
|
"Output CoNLL dataset file is saved to output directory.") |
|
|
|
with open(args.source_file, encoding="utf-8") as f: |
|
annotations = json.load(f) |
|
|
|
print("Starting documents processing...") |
|
|
|
with open(os.path.join(args.output_dir, "poner.conll"), "w", encoding="utf-8") as f: |
|
for page in annotations: |
|
page_text_path = page["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() |
|
processed_page = process_text_annotations(page_text, page["ner"]) |
|
f.write(processed_page + "\n\n") |
|
|
|
print("Annotations are processed.") |
|
|
|
|
|
if __name__ == '__main__': |
|
main() |
|
|
