app.py

# importing libraries

import streamlit as st
import tensorflow as tf
import nltk
from PIL import Image
from nltk.stem import WordNetLemmatizer
from nltk.tokenize import RegexpTokenizer
import string
import re
import pandas as pd
import numpy as np


try:
    nltk.find("corpora/wordnet.zip")
except:
    nltk.download("wordnet")

# read files
try:
    acronyms_dict,contractions_dict,stops
except(NameError):
    acronyms_dict = pd.read_json("acronym.json", typ="series")
    contractions_dict = pd.read_json("contraction.json", typ="series")
    stops = list(pd.read_csv("stop_words.csv").values.flatten())





# preprocess function

# Defining tokenizer
regexp = RegexpTokenizer("[\w']+")

def preprocess(text):
    
    text = text.lower()                                                                                        # lowercase
    text = text.strip()                                                                                        # whitespaces
    
    # Removing html tags
    html = re.compile(r'<.*?>')
    text = html.sub(r'', text)                                                                                 # html tags
    
    # Removing emoji patterns
    emoji_pattern = re.compile("["
                           u"\U0001F600-\U0001F64F"  # emoticons
                           u"\U0001F300-\U0001F5FF"  # symbols & pictographs
                           u"\U0001F680-\U0001F6FF"  # transport & map symbols
                           u"\U0001F1E0-\U0001F1FF"  # flags (iOS)
                           u"\U00002702-\U000027B0"
                           u"\U000024C2-\U0001F251"
                           "]+", flags = re.UNICODE)
    text = emoji_pattern.sub(r'', text)                                                                         # unicode char
    
    # Removing urls
    http = "https?://\S+|www\.\S+" # matching strings beginning with http (but not just "http")
    pattern = r"({})".format(http) # creating pattern
    text = re.sub(pattern, "", text)                                                                            # remove urls
    
    # Removing twitter usernames
    pattern = r'@[\w_]+'
    text = re.sub(pattern, "", text)                                                                            # remove @twitter usernames
    
    # Removing punctuations and numbers
    punct_str = string.punctuation + string.digits
    punct_str = punct_str.replace("'", "")
    punct_str = punct_str.replace("-", "")
    text = text.translate(str.maketrans('', '', punct_str))                                                     # punctuation and numbers
    
    # Replacing "-" in text with empty space
    text = text.replace("-", " ")                                                                               # "-"
    
    # Substituting acronyms
    words = []
    for word in regexp.tokenize(text):
        if word in acronyms_dict.index:
            words = words + acronyms_dict[word].split()
        else:
            words = words + word.split()
    text = ' '.join(words)                                                                                       # acronyms
    
    # Substituting Contractions
    words = []
    for word in regexp.tokenize(text):
        if word in contractions_dict.index:
            words = words + contractions_dict[word].split()
        else:
            words = words + word.split()
    text = " ".join(words)                                                                                       # contractions
    
    punct_str = string.punctuation
    text = text.translate(str.maketrans('', '', punct_str))                                                     # punctuation again to remove "'"                                                                        # spellchecker
    

    # lemmatization
    lemmatizer = WordNetLemmatizer()
    text = " ".join([lemmatizer.lemmatize(word) for word in regexp.tokenize(text)])                             # lemmatize
    
    # Stopwords Removal
    text = ' '.join([word for word in regexp.tokenize(text) if word not in stops])                              # stopwords
    
    # Removing all characters except alphabets and " " (space)
    filter = string.ascii_letters + " "
    text = "".join([chr for chr in text if chr in filter])                                                      # remove all characters except alphabets and " " (space)
    
    # Removing words with one alphabet occuring more than 3 times continuously
    pattern = r'\b\w*?(.)\1{2,}\w*\b'
    text = re.sub(pattern, "", text).strip()                                                                    # remove words with one alphabet occuring more than 3 times continuously
    
    # Removing words with less than 3 characters
    short_words = r'\b\w{1,2}\b'
    text = re.sub(short_words, "", text)                                                                     # remove words with less than 3 characters
    
    # return final output
    return text



# making frontend
st.write("# Disaster Tweet Prediction")

img=Image.open("disaster.jpg")
st.image(img,width=500)


tweet=st.text_input(label="",value="Enter or paste your tweet here")




# model load
@st.cache_resource
def cache_model(model_name):
    model=tf.keras.models.load_model(model_name)
    return model

model=cache_model("tweet_model")



if len(tweet)>0:
    clean_tweet=preprocess(tweet)
    y_pred=model.predict([clean_tweet])
    y_pred_num=int(np.round(y_pred)[0][0])


    if y_pred_num==0:
        st.write(f"## Non-Disaster tweet with disaster probability {round(y_pred[0][0],4)*100}%")

    else:
        st.write(f"## Disaster tweet with disaster probability {round(y_pred[0][0],4)*100}%")