SanjanaSuryadevara/Sentiment_Analysis_using_Word2Vector

1# ---------------------------------------------------------2# 🎬 Sentiment Analysis App (Word2Vec)3# ---------------------------------------------------------4import streamlit as st5import pandas as pd6import numpy as np7import re8import nltk9from nltk.stem import WordNetLemmatizer10from nltk.corpus import stopwords11from sklearn.model_selection import train_test_split12from sklearn.metrics import accuracy_score, confusion_matrix, classification_report13from sklearn.ensemble import RandomForestClassifier14from gensim.models import Word2Vec15import matplotlib.pyplot as plt16import seaborn as sns17 18# Optional: SMOTE for imbalance19from imblearn.over_sampling import SMOTE20 21# ---------------------------------------------------------22# Download NLTK Resources23# ---------------------------------------------------------24nltk.download('wordnet')25nltk.download('omw-1.4')26nltk.download('stopwords')27 28stop_words = set(stopwords.words('english'))29 30# ---------------------------------------------------------31# Page Design & Custom CSS32# ---------------------------------------------------------33def add_custom_styles():34    st.markdown(35        f"""36        <style>37        .stApp {{38            background-image: 39              linear-gradient(rgba(255, 255, 255, 0.85), rgba(255, 255, 255, 0.85)),40              url("https://images.unsplash.com/photo-1506744038136-46273834b3fb?auto=format&fit=crop&w=1470&q=80");41            background-attachment: fixed;42            background-size: cover;43            background-position: center;44        }}45        textarea {{46            font-size: 20px !important;47            font-weight: 900 !important;48            color: black !important;49        }}50        textarea::placeholder {{51            color: black !important;52            font-weight: 900 !important;53            font-size: 20px !important;54        }}55        </style>56        """,57        unsafe_allow_html=True58    )59 60add_custom_styles()61 62# ---------------------------------------------------------63# Data Loading & Preprocessing64# ---------------------------------------------------------65@st.cache_data66def load_and_preprocess():67    df = pd.read_csv("comments_with_sentiment.csv")68    69    # Remove rows where 'Comment' is missing or empty70    df['Comment'] = df['Comment'].astype(str).str.strip()71    df = df[df['Comment'] != '']72 73    # Check if Sentiment column exists74    if 'Sentiment Analysis' not in df.columns:75        raise KeyError("Column 'Sentiment Analysis' not found in CSV!")76    df['Sentiment Analysis'] = df['Sentiment Analysis'].fillna('Neutral')77    78    # Drop duplicate comments79    df = df.drop_duplicates(subset=['Comment'])80    81    # Clean text82    df['Comment'] = df['Comment'].str.lower()83    df['Comment'] = df['Comment'].apply(lambda x: re.sub(r'[^a-z\s]', '', x)).str.strip()84    85    # Tokenize & Lemmatize, remove stopwords86    lemmatizer = WordNetLemmatizer()87    df['tokens'] = df['Comment'].apply(88        lambda x: [lemmatizer.lemmatize(word) for word in x.split() if word not in stop_words]89    )90    91    return df92 93# ---------------------------------------------------------94# Custom Word2Vec Transformer95# ---------------------------------------------------------96class Word2VecVectorizer:97    def __init__(self, vector_size=100, window=5, min_count=2, sg=1):98        self.vector_size = vector_size99        self.window = window100        self.min_count = min_count101        self.sg = sg102        self.w2v_model = None103 104    def fit(self, texts):105        self.w2v_model = Word2Vec(106            sentences=texts,107            vector_size=self.vector_size,108            window=self.window,109            min_count=self.min_count,110            sg=self.sg,111            workers=4112        )113        return self114 115    def transform(self, texts):116        def document_vector(tokens):117            tokens = [t for t in tokens if t in self.w2v_model.wv]118            if len(tokens) == 0:119                return np.random.normal(size=self.vector_size)120            return np.mean(self.w2v_model.wv[tokens], axis=0)121        return np.array([document_vector(t) for t in texts])122 123# ---------------------------------------------------------124# Train Model Pipeline125# ---------------------------------------------------------126@st.cache_resource127def train_pipeline(df):128    X = df['tokens']129    y = df['Sentiment Analysis']130    131    X_train, X_test, y_train, y_test = train_test_split(132        X, y, test_size=0.2, random_state=42, stratify=y133    )134 135    w2v_vectorizer = Word2VecVectorizer(vector_size=100, window=5, min_count=2, sg=1)136    w2v_vectorizer.fit(X_train)137 138    X_train_vec = w2v_vectorizer.transform(X_train)139    X_test_vec = w2v_vectorizer.transform(X_test)140 141    # SMOTE to handle imbalance142    smote = SMOTE(random_state=42)143    X_train_vec, y_train = smote.fit_resample(X_train_vec, y_train)144 145    clf = RandomForestClassifier(146        n_estimators=500,147        max_depth=15,148        random_state=42,149        class_weight='balanced'150    )151    clf.fit(X_train_vec, y_train)152    y_pred = clf.predict(X_test_vec)153 154    acc = accuracy_score(y_test, y_pred)155    return w2v_vectorizer, clf, acc156 157# ---------------------------------------------------------158# Streamlit App UI159# ---------------------------------------------------------160st.markdown("## 🎬✨ **Sentiment Analysis** ✨🎬")161st.markdown("##### 💡 Enter a comment to analyze its sentiment using Word2Vec")162 163# Load & preprocess data164df = load_and_preprocess()165 166# Train model pipeline167vectorizer, model, acc = train_pipeline(df)168st.success(f"✅ Model trained successfully with Accuracy: **{acc:.2f}**")169 170# User input171user_input = st.text_area("💬 Enter your comment here:", key="user_input")172 173if st.button("🧠 Predict Sentiment"):174    lemmatizer = WordNetLemmatizer()175    cleaned = re.sub(r'[^a-z\s]', '', user_input.lower()).strip()176    tokens = [lemmatizer.lemmatize(word) for word in cleaned.split() if word not in stop_words]177    vec = vectorizer.transform([tokens])178    prediction = model.predict(vec)[0]179 180    # Dynamic button color181    color = "#4CAF50"  # Green for Positive182    if prediction.lower() == "neutral":183        color = "#2196F3"  # Blue184    elif prediction.lower() == "negative":185        color = "#F44336"  # Red186 187    st.markdown(188        f"""189        <div style="padding: 10px; text-align: center; background-color: {color}; 190                    color: white; font-size: 22px; font-weight: 900; border-radius: 10px;">191            🎯 Predicted Sentiment: {prediction}192        </div>193        """,194        unsafe_allow_html=True195    )196