app.py

import subprocess

# Install hunspell and its dependencies, pip wheels are completely broken
import subprocess

# Execute the shell script
subprocess.call(['sh', 'install_hunspell.sh'])

# Import hunspell
import hunspell
# Main imports
import gradio as gr
import re
import stanza
import spacy
import pandas as pd

def create_settlement_and_country_lists():
    settlement_list = []
    country_list = []

    # Read Ukrainian settlement names from CSV file
    df_settlements = pd.read_csv("assets/locations/ukrainian_settlement_mames.csv", encoding="utf-8")
    ukrainian_settlements = df_settlements["Назва об'єкта українською мовою"].values.tolist()
    settlement_list.extend(ukrainian_settlements)

    # Read European settlement names from CSV file
    df_eu_settlements = pd.read_csv("assets/locations/european_cities.csv", encoding="utf-8")
    european_settlements = df_eu_settlements["City"].values.tolist()
    settlement_list.extend(european_settlements)

    # Convert settlement list to lowercase
    settlement_list = [word.lower() for word in settlement_list]

    # Read country names from text file
    with open("assets/locations/countries.txt", "r", encoding="utf-8") as country_file:
        country_list = [line.strip().lower() for line in country_file]

    return settlement_list, country_list

# Call the function to create settlement and country lists
settlement_list, country_list = create_settlement_and_country_lists()

spellchecker = hunspell.HunSpell('assets/dictionaries/uk_UA.dic', 'assets/dictionaries/uk_UA.aff')

settlement_list = [s.lower() for s in settlement_list]  # Convert settlement list to lowercase
country_list = [c.lower() for c in country_list]  # Convert country list to lowercase

# Initialize Stanza NLP
stanza.download('uk')
nlp_stanza = stanza.Pipeline('uk', processors='tokenize,pos,ner')

# Load SpaCy NER model
nlp_spacy = spacy.load("uk_ner_web_trf_base")

def process_text_with_stanza(text):
    doc = nlp_stanza(text)
    return format_output(process_text(doc))

def process_text_with_spacy(text):
    doc = nlp_spacy(text)
    return format_output(process_text_spacy(doc))

def format_output(matches):
    formatted_matches = []
    for match in matches:
        location_type = match[0]
        entity = match[1]
        formatted_matches.append(f"{location_type}: {entity}")
    return "\n".join(formatted_matches) if formatted_matches else notify_no_result()

def notify_no_result():
    return "No locations found in the text."

def process_text(doc):
    starting_point_patterns = [r'(з|із|із-за|від|от|од){pos:IN} (\w+{ner:LOC})']
    destination_patterns = [r'(до|в|у|ув|к){pos:IN} (\w+{ner:LOC})']

    starting_point_matches = []
    for pattern in starting_point_patterns:
        matches = re.findall(pattern, doc.text)
        starting_point_matches.extend(matches)

    destination_matches = []
    for pattern in destination_patterns:
        matches = re.findall(pattern, doc.text)
        destination_matches.extend(matches)

    loc_entities = [ent.text for ent in doc.ents if ent.type == 'LOC']
    if len(loc_entities) == 2 and not starting_point_matches and not destination_matches:
        starting_point = loc_entities[0]
        destination = loc_entities[1]
        return [
            (starting_point, 'Starting Point', get_base_form_regex(starting_point, settlement_list, country_list, doc)),
            (destination, 'Destination', get_base_form_regex(destination, settlement_list, country_list, doc))
        ]

    if len(loc_entities) == 1 and not starting_point_matches and not destination_matches:
        return [(loc_entities[0], 'Unknown', get_base_form_regex(loc_entities[0], settlement_list, country_list, doc))]

    treated_matches = [
        (match[1], 'Starting Point', get_base_form_regex(match[1], settlement_list, country_list, doc))
        for match in starting_point_matches
    ] + [
        (match[1], 'Destination', get_base_form_regex(match[1], settlement_list, country_list, doc))
        for match in destination_matches
    ]

    formatted_matches = []
    for match in treated_matches:
        location_type = match[1]
        lemma_results = match[2][0]  # Access the first element of the nested list
        formatted_lemma = lemma_results[1].capitalize().strip('\n')
        formatted_matches.append((location_type, lemma_results[0], formatted_lemma))

    return formatted_matches

def process_text_spacy(doc):
    starting_point_patterns = [
        r'(з|із|із-за|від|от|од){pos:ADP} (\w+{ner:LOC})',
        r'(\w+{ner:LOC})\s+(з|із|із-за|від|от|од){pos:ADP}'
    ]
    destination_patterns = [
        r'(до|в|у|ув|к){pos:ADP} (\w+{ner:LOC})',
        r'(\w+{ner:LOC})\s+(до|в|у|ув|к){pos:ADP}'
    ]

    starting_point_matches = []
    for pattern in starting_point_patterns:
        matches = re.findall(pattern, doc.text)
        starting_point_matches.extend(matches)

    destination_matches = []
    for pattern in destination_patterns:
        matches = re.findall(pattern, doc.text)
        destination_matches.extend(matches)

    loc_entities = [ent.text for ent in doc.ents if ent.label_ == 'LOC']
    if len(loc_entities) == 2 and not starting_point_matches and not destination_matches:
        starting_point = loc_entities[0]
        destination = loc_entities[1]
        return [
            (starting_point, 'Starting Point', get_base_form_stanza(starting_point, settlement_list, country_list, doc)),
            (destination, 'Destination', get_base_form_stanza(destination, settlement_list, country_list, doc))
        ]

    if len(loc_entities) == 1 and not starting_point_matches and not destination_matches:
        return [(loc_entities[0], 'Unknown', get_base_form_stanza(loc_entities[0], settlement_list, country_list, doc))]

    treated_matches = [
        (match[1], 'Starting Point', get_base_form_stanza(match[1], settlement_list, country_list, doc))
        for match in starting_point_matches
    ] + [
        (match[1], 'Destination', get_base_form_stanza(match[1], settlement_list, country_list, doc))
        for match in destination_matches
    ]

    formatted_matches = []
    for match in treated_matches:
        location_type = match[1]
        lemma_results = match[2]  # Use directly, as it's already the required format
        formatted_lemma = lemma_results.capitalize().strip('\n')
        formatted_matches.append((location_type, lemma_results, formatted_lemma))

    return formatted_matches

def get_base_form_stanza(word, settlement_list, country_list, doc):
    token = None
    base_form = ""
    for sent in doc.sentences:
        for wrd in sent.words:
            if wrd.text.lower() == word.lower():
                token = wrd
                break

    if token is not None:
        if token.upos == 'PROPN' and token.text.lower() not in settlement_list and token.text.lower() not in country_list:
            base_form = token.lemma
        else:
            base_form = token.text

    return base_form

def get_base_form_regex(word, settlement_list, country_list, doc):
    base_form = ""
    base_form_regex = ""

    if word.lower() in settlement_list or word.lower() in country_list:
        base_form = word.lower()
    else:
        base_form = get_base_form_stanza(word, settlement_list, country_list, doc)

    if base_form:
        base_form_regex = base_form

    return base_form_regex, base_form

iface = gr.Interface(
    fn=[process_text_with_stanza, process_text_with_spacy],
    inputs=gr.inputs.Textbox(lines=5, label="Input Text"),
    outputs=["text", "text"],
    title="Text Processing Demo",
    description="A demo to process text and extract locations using Stanza and SpaCy.",
    examples=[
        ["Автобус з Києва до Житомира"],
        ["Автобус з Києва в Бердичів"],
        ["Поїздка з Варшави до Івано-Франківська"],
    ]
)

iface.launch()