from flask import Flask, request, jsonify, send_from_directory import pandas as pd import torch from transformers import BertTokenizer, BertForSequenceClassification from wordcloud import WordCloud import uuid import io import base64 import os from PIL import Image app = Flask(__name__) UPLOAD_FOLDER = "uploads" # os.makedirs(UPLOAD_FOLDER, exist_ok=True) @app.route('/uploads/') def uploaded_file(filename): return send_from_directory(app.config['UPLOAD_FOLDER'], filename) # Load model and tokenizer once tokenizer = BertTokenizer.from_pretrained("bert-base-multilingual-cased") model_path = "./src/emotion_final_model" device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = BertForSequenceClassification.from_pretrained(model_path).to(device) model.eval() # Label Mapping label_mapping = {0: "negative", 1: "neutral", 2: "positive"} @app.route('/predict', methods=['POST']) def predict(): data = request.get_json() text = data.get('text') if not text: return jsonify({"error": "No text provided"}), 400 inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512) inputs = {key: value.to(device) for key, value in inputs.items()} with torch.no_grad(): outputs = model(**inputs) logits = outputs.logits predicted_class_idx = torch.argmax(logits, dim=-1).item() sentiment = label_mapping[predicted_class_idx] return jsonify({"sentiment": sentiment}) aspect_keywords = { "Quality": ["quality", "material", "durable", "performance", "sturdy", "broken", "defective", "معیار", "ٹوٹا ہوا", "خراب"], "Price": ["price", "cheap", "expensive", "value", "cost", "قیمت", "مہنگا", "سستا", "قیمت زیادہ"], "Delivery": ["delivery", "shipping", "arrived", "late", "courier", "ترسیل", "شپنگ", "تاخیر", "دیر سے پہنچا"], "Usability": ["easy to use", "setup", "installation", "instructions", "user-friendly", "آسان", "استعمال میں آسان", "سیٹ اپ", "تنصیب"], "Design": ["design", "style", "appearance", "color", "looks", "ڈیزائن", "خوبصورتی", "رنگ", "ساخت"], "Warranty/Support": ["warranty", "support", "return", "replacement", "service center", "وارنٹی", "واپسی", "تبادلہ", "سروس سینٹر"] } def detect_aspects(text): text_lower = text.lower() detected = [] for aspect, keywords in aspect_keywords.items(): if any(keyword in text_lower for keyword in keywords): detected.append(aspect) return detected @app.route("/analyze", methods=["POST"]) def analyze(): if 'file' not in request.files: return jsonify({"error": "No file uploaded"}), 400 file = request.files['file'] print(file.filename) df = pd.read_csv(file) print(df.to_string()) total_positive = 0 total_negative = 0 total_neutral = 0 all_text = "" # Aspect summary aspect_summary = {aspect: {"positive": 0, "negative": 0, "neutral": 0, "total": 0} for aspect in aspect_keywords} for text in df['Review'].dropna(): inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512) inputs = {k: v.to(device) for k, v in inputs.items()} with torch.no_grad(): outputs = model(**inputs) predicted_class_idx = torch.argmax(outputs.logits, dim=-1).item() sentiment = label_mapping[predicted_class_idx] if sentiment == "positive": total_positive += 1 elif sentiment == "negative": total_negative += 1 else: total_neutral += 1 all_text += " " + text detected_aspects = detect_aspects(text) for aspect in detected_aspects: aspect_summary[aspect][sentiment] += 1 aspect_summary[aspect]["total"] += 1 # Generate WordCloud wordcloud = WordCloud(width=800, height=400, background_color='white', font_path='src/urdu_font.ttf').generate(all_text) # Save in uploads folder # if not os.path.exists("uploads"): # os.makedirs("uploads") wordcloud_path = os.path.join("uploads", f"wordcloud{uuid.uuid4()}.png") wordcloud.to_file(wordcloud_path) # Convert image to base64 with open(wordcloud_path, "rb") as image_file: encoded_image = base64.b64encode(image_file.read()).decode('utf-8') print({ "total_positive": total_positive, "total_negative": total_negative, "total_neutral": total_neutral, "aspect_summary": aspect_summary, "wordcloud_image_path": wordcloud_path, }) return jsonify({ "total_positive": total_positive, "total_negative": total_negative, "total_neutral": total_neutral, "aspect_summary": aspect_summary, "wordcloud_image_path": wordcloud_path, # "wordcloud_image_base64": encoded_image }) def run_flask(): app.run(host="0.0.0.0", port=5000) import threading threading.Thread(target=run_flask).start() import streamlit as st import pandas as pd import plotly.express as px from io import BytesIO, StringIO from PIL import Image import random import requests import os import uuid import tempfile API_URL = 'http://127.0.0.1:5000/analyze' # ------------------- # PAGE CONFIG & THEME # ------------------- st.set_page_config( page_title="Multilingual Sentiment Analyzer", layout="wide" ) st.markdown(""" """, unsafe_allow_html=True) st.markdown("""

🌍 Multilingual Sentiment Analysis Dashboard

Upload a CSV to explore sentiment Report. With sentiment analysis, you can catch early signals, reduce risk, and validate market fit — even across global audiences.

""", unsafe_allow_html=True) # ------------------- # DUMMY DATA FUNCTION # ------------------- def load_dummy_data(): return pd.DataFrame({ "Review": [ "La livraison était très rapide et le service excellent.", "The product quality was terrible, I want a refund.", "Servicio al cliente fue amable pero no resolvieron mi problema.", "Das Produkt kam beschädigt an und der Support war unhöflich.", "Great value for the price, I'm very happy!", "Muy mal embalaje, pero el envío fue rápido.", "客服很好，但产品描述不准确。", "Perfect fit, just as described. Will buy again!" ] }) # ------------------- # MAIN UPLOAD BLOCK (VISIBLE) # ------------------- with st.expander("📁 Upload Your CSV File", expanded=True): uploaded_file = st.file_uploader("Choose a CSV file with reviews", type=["csv"]) # Analysis button moved here, right after file upload run_analysis = st.button("🚀 Run Analysis", type="primary") # Load Data: Uploaded or Dummy if uploaded_file: try: # Read uploaded CSV file df = pd.read_csv(uploaded_file) if df.empty: st.error("The uploaded CSV file is empty.") df = load_dummy_data() else: st.success("✅ File uploaded successfully!") except Exception as e: st.error(f"Error reading CSV: {e}") df = load_dummy_data() else: st.info("Using built-in demo data. Upload a CSV to use your own.") df = load_dummy_data() # Preview data st.write("✅ App is running! Here's a sample:") st.dataframe(df.head()) # Column selection # text_column = st.selectbox("📝 Select the column with review text:", df.columns) # enable_aspect = st.checkbox("🔍 Include Aspect Report (Optional)", value=True) text_column = 'Review' enable_aspect = True # ------------------- # SENTIMENT METRICS # ------------------- st.markdown("---") st.markdown("### 🔎 Sentiment Analysis Results") # MOCK SENTIMENT PREDICTION def fake_sentiment_predict(text): return random.choice(["Positive", "Negative", "Neutral"]), round(random.uniform(0.65, 0.99), 2) # Store the analyzed dataframe in session state if 'analyzed_df' not in st.session_state: st.session_state.analyzed_df = df.copy() # Initialize variables positive, negative, neutral, total = 0, 0, 0, 0 # Process the data when the Run Analysis button is clicked if run_analysis: if not enable_aspect: # Use fake predictions if not calling the API fake_results = [fake_sentiment_predict(text) for text in df[text_column]] sentiments, confidences = zip(*fake_results) # Update the analyzed dataframe st.session_state.analyzed_df = df.copy() st.session_state.analyzed_df["Sentiment"] = sentiments st.session_state.analyzed_df["Confidence"] = confidences # Count sentiment sentiment_counts = pd.Series(sentiments).value_counts() positive = sentiment_counts.get("Positive", 0) negative = sentiment_counts.get("Negative", 0) neutral = sentiment_counts.get("Neutral", 0) total = positive + negative + neutral else: # API Call will be made - handled below pass def percent(part): return f"{round((part / total) * 100)}%" if total else "0%" # Layout: Cards + Chart card_col, chart_col = st.columns([1.2, 2]) with card_col: st.markdown("""
""", unsafe_allow_html=True) if total > 0: # Only display chart if we have data with chart_col: fig = px.pie( names=["Positive", "Negative", "Neutral"], values=[positive, negative, neutral], color_discrete_map={ "Positive": "#66bb6a", "Negative": "#ef5350", "Neutral": "#42a5f5" } ) fig.update_traces( textinfo='percent+label', hoverinfo='label+percent+value', pull=[0.03, 0.03, 0.03] ) fig.update_layout( margin=dict(t=20, b=20, l=10, r=10), paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)", font_color="white" ) st.plotly_chart(fig, use_container_width=True) # ------------------- # ASPECT REPORT (API Call) # ------------------- if run_analysis and enable_aspect: st.subheader("📌 Aspect Sentiment Summary") with st.spinner("Fetching Aspect Report from API..."): try: # Create a dataframe with only the text column api_df = df.copy() # Create a unique filename with UUID unique_filename = f"temp_reviews_{uuid.uuid4()}.csv" # Write to a physical temporary file with UUID in name with tempfile.NamedTemporaryFile(delete=False, suffix=unique_filename) as tmp: api_df.to_csv(tmp.name, index=False) tmp_file_path = tmp.name # Open the file in binary mode with open(tmp_file_path, 'rb') as file: # Send the actual file files = {'file': (unique_filename, file, 'text/csv')} response = requests.post(API_URL, files=files) # Clean up temporary file try: os.unlink(tmp_file_path) except: pass # Silently fail if we can't delete the temp file if response.status_code == 200: response_json = response.json() # Store API response in session state st.session_state.api_response_json = response_json # Update metrics from API response positive = response_json.get("total_positive", 0) negative = response_json.get("total_negative", 0) neutral = response_json.get("total_neutral", 0) total = positive + negative + neutral # Calculate percentages for Excel report if not provided by API if "positive_percentage" not in response_json and total > 0: response_json["positive_percentage"] = round((positive / total) * 100) response_json["negative_percentage"] = round((negative / total) * 100) response_json["neutral_percentage"] = round((neutral / total) * 100) # Add total reviews to response_json if not present if "total_reviews" not in response_json: response_json["total_reviews"] = total # Update metrics from API response positive = response_json.get("total_positive", 0) negative = response_json.get("total_negative", 0) neutral = response_json.get("total_neutral", 0) total = positive + negative + neutral # Update the metrics cards with new data with card_col: st.markdown(f"""
📊 Total Reviews {total}
✅ {positive} Positive {percent(positive)} of total
❗ {negative} Negative {percent(negative)} of total
😐 {neutral} Neutral {percent(neutral)} of total
""", unsafe_allow_html=True) # Update the pie chart with chart_col: fig = px.pie( names=["Positive", "Negative", "Neutral"], values=[positive, negative, neutral], color_discrete_map={ "Positive": "#66bb6a", "Negative": "#ef5350", "Neutral": "#42a5f5" } ) fig.update_traces( textinfo='percent+label', hoverinfo='label+percent+value', pull=[0.03, 0.03, 0.03] ) fig.update_layout( margin=dict(t=20, b=20, l=10, r=10), paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)", font_color="white" ) st.plotly_chart(fig, use_container_width=True) # Update the analyzed dataframe with sentiment results from API if "review_details" in response_json: # Create a new dataframe from the API results api_result_df = pd.DataFrame(response_json["review_details"]) # Store it in session state st.session_state.analyzed_df = api_result_df else: # If review_details not provided, create basic sentiment columns st.session_state.analyzed_df = df.copy() # Try to extract sentiments if available in the API response if "sentiments" in response_json: st.session_state.analyzed_df["Sentiment"] = response_json["sentiments"] # Add any other available result fields for key in ["confidences", "languages"]: if key in response_json: column_name = key.rstrip("s").capitalize() # Convert "confidences" to "Confidence" st.session_state.analyzed_df[column_name] = response_json[key] # Prepare aspect DataFrame aspect_rows = [] for aspect, values in response_json["aspect_summary"].items(): aspect_rows.append({ "Aspect": aspect, "Positive": values["positive"], "Negative": values["negative"], "Neutral": values["neutral"], "Total": values["total"] }) aspect_df = pd.DataFrame(aspect_rows) # Display aspect data if we have any if not aspect_df.empty and aspect_df["Total"].sum() > 0: # Store aspect dataframe in session state st.session_state.aspect_dataframe = aspect_df st.dataframe(aspect_df) # Prepare data for bar chart melted = aspect_df.melt( id_vars="Aspect", value_vars=["Positive", "Negative", "Neutral"], var_name="Sentiment", value_name="Count" ) col1, col2 = st.columns([4, 2]) with col1: st.markdown("### 📊 Sentiment by Aspect") bar_chart = px.bar( melted, x="Aspect", y="Count", color="Sentiment", barmode="group", title=None, color_discrete_map={ "Positive": "#66bb6a", "Negative": "#ef5350", "Neutral": "#42a5f5" } ) # Update chart theme for dark mode bar_chart.update_layout( paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)", font_color="white", xaxis=dict(gridcolor="rgba(255,255,255,0.1)"), yaxis=dict(gridcolor="rgba(255,255,255,0.1)") ) st.plotly_chart(bar_chart, use_container_width=True) with col2: st.markdown("### 🌀 Review Keywords") # Try to display wordcloud from API if "wordcloud_image_base64" in response_json: import base64 st.markdown("
", unsafe_allow_html=True) st.image( BytesIO(base64.b64decode(response_json["wordcloud_image_base64"])), caption="Keyword Cloud", use_container_width=True ) else: try: # Try local wordcloud file as fallback wordcloud_path = response_json.get("wordcloud_image_path") if wordcloud_path and os.path.exists(wordcloud_path): image = Image.open(wordcloud_path) st.markdown("
", unsafe_allow_html=True) st.image(image, caption="Keywords", use_container_width=True) else: # Try default wordcloud if os.path.exists("wordcloud.jpg"): image = Image.open("wordcloud.jpg") st.markdown("
", unsafe_allow_html=True) st.image(image, caption="Keywords", use_container_width=True) except Exception as e: st.warning(f"⚠ Word cloud image not found: {e}") else: st.info("No aspects detected in the reviews.") else: st.error(f"API Error: {response.status_code} - {response.text}") except Exception as e: st.error(f"API call failed: {e}") import traceback st.code(traceback.format_exc(), language="python") # ------------------- # DOWNLOAD BUTTON # ------------------- if run_analysis or total > 0: st.subheader("📥 Download Analyzed File") def generate_excel_report(df, aspect_data=None, response_json=None): output = BytesIO() with pd.ExcelWriter(output, engine='xlsxwriter') as writer: # Sheet 1: Main sentiment results df.to_excel(writer, index=False, sheet_name='Sentiment_Report') # Sheet 2: Aspect analysis (if available) if aspect_data is not None and not aspect_data.empty: aspect_data.to_excel(writer, index=False, sheet_name='Aspect_Analysis') # Sheet 3: Summary stats (if API response available) if response_json: # Create a summary dataframe summary_data = { 'Metric': ['Total Reviews', 'Positive', 'Negative', 'Neutral'], 'Count': [ response_json.get('total_reviews', 0), response_json.get('total_positive', 0), response_json.get('total_negative', 0), response_json.get('total_neutral', 0) ], 'Percentage': [ '100%', f"{response_json.get('positive_percentage', 0)}%", f"{response_json.get('negative_percentage', 0)}%", f"{response_json.get('neutral_percentage', 0)}%" ] } summary_df = pd.DataFrame(summary_data) summary_df.to_excel(writer, index=False, sheet_name='Summary') # Add any other relevant data from API response if 'review_details' in response_json: details_df = pd.DataFrame(response_json['review_details']) details_df.to_excel(writer, index=False, sheet_name='Review_Details') # Get workbook and add some formatting workbook = writer.book # Add formatting header_format = workbook.add_format({ 'bold': True, 'text_wrap': True, 'valign': 'top', 'border': 1 }) # Apply formatting to each worksheet safely for sheet_name in writer.sheets: worksheet = writer.sheets[sheet_name] # Get column names from the DataFrame based on sheet name if sheet_name == 'Sentiment_Report': columns = df.columns elif sheet_name == 'Aspect_Analysis' and aspect_data is not None: columns = aspect_data.columns elif sheet_name == 'Summary': columns = summary_data.keys() elif sheet_name == 'Review_Details' and 'review_details' in response_json: columns = details_df.columns else: continue # Write headers with formatting for col_num, value in enumerate(columns): worksheet.write(0, col_num, value, header_format) # Auto-adjust columns' width (supported in newer versions) try: worksheet.autofit() except AttributeError: # Fallback for older xlsxwriter versions for col_num, value in enumerate(columns): # Set width based on header content worksheet.set_column(col_num, col_num, max(10, len(str(value)) + 2)) return output.getvalue() # Store API response in session state to access it for download if 'api_response_json' not in st.session_state: st.session_state.api_response_json = None if 'aspect_dataframe' not in st.session_state: st.session_state.aspect_dataframe = None # Update these values when API response is received if run_analysis and enable_aspect and 'response_json' in locals(): st.session_state.api_response_json = response_json if 'aspect_df' in locals() and not aspect_df.empty: st.session_state.aspect_dataframe = aspect_df st.download_button( label="📥 Download Results as Excel", data=generate_excel_report( st.session_state.analyzed_df, # Use the analyzed dataframe instead of original df st.session_state.aspect_dataframe, st.session_state.api_response_json ), file_name="sentiment_analysis_report.xlsx", mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet" ) # Add a footer with dark theme st.markdown("""

Multilingual Sentiment Analysis Dashboard | Made with Streamlit

""", unsafe_allow_html=True)