app.py

import streamlit as st
import numpy as np
import transformers
import re
import string
import preprocessor as pre

import torch
from transformers import BertTokenizer, BertForSequenceClassification

with open("style.css") as f:
    st.markdown('<style>{}</style>'.format(f.read()), unsafe_allow_html=True) 

# Preparation model and tokenizer
model_path = "ninahf1503/SA-BERTchatgptapp"
tokenizer = BertTokenizer.from_pretrained(model_path)
model = BertForSequenceClassification.from_pretrained(model_path, ignore_mismatched_sizes=True )

# Define the maximum sequence length
seq_max_length = 55

# Function to tokenizing input text
def tokenizing_text(sentence):
    sentence = preprocess_text(sentence)
    encoded = tokenizer.encode_plus(
        sentence,
        add_special_tokens=True,
        max_length=seq_max_length,
        truncation=True,
        padding='max_length',
        return_tensors='pt'
    )

    input_ids = encoded['input_ids']
    attention_mask = encoded['attention_mask']
    return input_ids, attention_mask

# Function to preprocessing input text
def preprocess_text(sentence):
    re_cleansing = "@\S+|https?:\S+|http?:\S|#[A-Za-z0-9]+|^RT[\s]+|(^|\W)\d+"
    for punctuation in string.punctuation:
        sentence = sentence.encode().decode('unicode_escape')
        sentence = re.sub(r'\n', ' ', sentence)
        sentence = pre.clean(sentence)
        sentence = re.sub(r'[^\w\s]', ' ', sentence)
        sentence = re.sub(r'[0-9]', ' ', sentence)
        sentence = re.sub(re_cleansing, ' ', sentence).strip()
        sentence = sentence.replace(punctuation, '')
        sentence = sentence.lower()
    return sentence

# Function to predict sentiment
def predict_sentiment(input_text):
    input_ids, attention_mask = tokenizing_text(input_text)

    with torch.no_grad():
        outputs = model(input_ids, attention_mask)
    
    logits = outputs.logits
    predict_class = torch.argmax(logits, dim=1).item()
    
    label_sentiment = {0: "Bad", 1: "Good", 2: "Neutral"}
    predict_label = label_sentiment[predict_class]
    
    return predict_label



# Streamlit web app
def main():
    st.title("Sentimen Analysis", anchor=False)
    tweet_text = st.text_area(" ", placeholder="Enter the sentence you want to analyze", label_visibility="collapsed")
    
    if st.button("SUBMIT"):
        if tweet_text.strip() == "":
            st.title("Text Input Still Empty", anchor=False)
            st.info("Please fill in the sentence you want to analyze")
        else:
            sentiment = predict_sentiment(tweet_text)
            if sentiment == "Good":
                st.title("Sentiment Analysis Results", anchor=False)
                st.markdown('<div style="background-color: #5d9c59; padding: 16px; border-radius: 5px; font-weight: bold; color:white;">This sentence contains a positive sentiment</div>', unsafe_allow_html=True)
            elif sentiment == "Bad":
                st.title("Sentiment Analysis Results", anchor=False)
                st.markdown('<div style="background-color: #df2e38; padding: 16px; border-radius: 5px; font-weight: bold; color:white;">This sentence contains a negative sentiment</div>', unsafe_allow_html=True)
            else:
                st.title("Sentiment Analysis Results", anchor=False)
                st.markdown('<div style="background-color: #ffa500; padding: 16px; border-radius: 5px; font-weight: bold; color:white;">This sentence is neutral</div>', unsafe_allow_html=True)

if __name__ == "__main__":
    main()