Upload 2 files
Browse files- app.py +326 -0
- requirements.txt +6 -3
app.py
ADDED
|
@@ -0,0 +1,326 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
import torch
|
| 3 |
+
from transformers import BertForSequenceClassification, BertTokenizerFast
|
| 4 |
+
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
| 5 |
+
import time
|
| 6 |
+
import pandas as pd
|
| 7 |
+
import base64
|
| 8 |
+
from PIL import Image
|
| 9 |
+
import io
|
| 10 |
+
|
| 11 |
+
# Set page configuration
|
| 12 |
+
st.set_page_config(
|
| 13 |
+
page_title="SMS Spam Guard",
|
| 14 |
+
page_icon="🛡️",
|
| 15 |
+
layout="wide",
|
| 16 |
+
initial_sidebar_state="expanded"
|
| 17 |
+
)
|
| 18 |
+
|
| 19 |
+
# Generate SafeTalk logo as base64 (blue shield with "ST" inside)
|
| 20 |
+
def create_logo():
|
| 21 |
+
from PIL import Image, ImageDraw, ImageFont
|
| 22 |
+
import io
|
| 23 |
+
import base64
|
| 24 |
+
|
| 25 |
+
# Create a new image with a transparent background
|
| 26 |
+
img = Image.new('RGBA', (200, 200), color=(0, 0, 0, 0))
|
| 27 |
+
draw = ImageDraw.Draw(img)
|
| 28 |
+
|
| 29 |
+
# Draw a shield shape
|
| 30 |
+
shield_color = (30, 58, 138) # Dark blue
|
| 31 |
+
|
| 32 |
+
# Shield outline
|
| 33 |
+
points = [(100, 10), (180, 50), (160, 170), (100, 190), (40, 170), (20, 50)]
|
| 34 |
+
draw.polygon(points, fill=shield_color)
|
| 35 |
+
|
| 36 |
+
# Try to load a font, or use default
|
| 37 |
+
try:
|
| 38 |
+
font = ImageFont.truetype("arial.ttf", 80)
|
| 39 |
+
except IOError:
|
| 40 |
+
font = ImageFont.load_default()
|
| 41 |
+
|
| 42 |
+
# Add "ST" text in white
|
| 43 |
+
draw.text((70, 60), "ST", fill=(255, 255, 255), font=font)
|
| 44 |
+
|
| 45 |
+
# Convert to base64 for embedding
|
| 46 |
+
buffered = io.BytesIO()
|
| 47 |
+
img.save(buffered, format="PNG")
|
| 48 |
+
return base64.b64encode(buffered.getvalue()).decode()
|
| 49 |
+
|
| 50 |
+
# Custom CSS for styling
|
| 51 |
+
st.markdown("""
|
| 52 |
+
<style>
|
| 53 |
+
.main-header {
|
| 54 |
+
font-size: 2.5rem !important;
|
| 55 |
+
color: #1E3A8A;
|
| 56 |
+
font-weight: 700;
|
| 57 |
+
margin-bottom: 0.5rem;
|
| 58 |
+
}
|
| 59 |
+
.sub-header {
|
| 60 |
+
font-size: 1.1rem;
|
| 61 |
+
color: #6B7280;
|
| 62 |
+
margin-bottom: 2rem;
|
| 63 |
+
}
|
| 64 |
+
.highlight {
|
| 65 |
+
background-color: #F3F4F6;
|
| 66 |
+
padding: 1.5rem;
|
| 67 |
+
border-radius: 0.5rem;
|
| 68 |
+
margin-bottom: 1rem;
|
| 69 |
+
}
|
| 70 |
+
.result-card {
|
| 71 |
+
background-color: #F0F9FF;
|
| 72 |
+
padding: 1.5rem;
|
| 73 |
+
border-radius: 0.5rem;
|
| 74 |
+
border-left: 5px solid #3B82F6;
|
| 75 |
+
margin-bottom: 1rem;
|
| 76 |
+
}
|
| 77 |
+
.spam-alert {
|
| 78 |
+
background-color: #FEF2F2;
|
| 79 |
+
border-left: 5px solid #EF4444;
|
| 80 |
+
}
|
| 81 |
+
.ham-alert {
|
| 82 |
+
background-color: #ECFDF5;
|
| 83 |
+
border-left: 5px solid #10B981;
|
| 84 |
+
}
|
| 85 |
+
.footer {
|
| 86 |
+
text-align: center;
|
| 87 |
+
margin-top: 3rem;
|
| 88 |
+
font-size: 0.8rem;
|
| 89 |
+
color: #9CA3AF;
|
| 90 |
+
}
|
| 91 |
+
.metrics-container {
|
| 92 |
+
display: flex;
|
| 93 |
+
justify-content: space-between;
|
| 94 |
+
margin-top: 1rem;
|
| 95 |
+
}
|
| 96 |
+
.metric-item {
|
| 97 |
+
text-align: center;
|
| 98 |
+
padding: 1rem;
|
| 99 |
+
background-color: #F9FAFB;
|
| 100 |
+
border-radius: 0.5rem;
|
| 101 |
+
box-shadow: 0 1px 3px rgba(0,0,0,0.1);
|
| 102 |
+
}
|
| 103 |
+
.language-tag {
|
| 104 |
+
display: inline-block;
|
| 105 |
+
padding: 0.25rem 0.5rem;
|
| 106 |
+
background-color: #E0E7FF;
|
| 107 |
+
color: #4F46E5;
|
| 108 |
+
border-radius: 9999px;
|
| 109 |
+
font-size: 0.8rem;
|
| 110 |
+
font-weight: 500;
|
| 111 |
+
margin-right: 0.5rem;
|
| 112 |
+
}
|
| 113 |
+
</style>
|
| 114 |
+
""", unsafe_allow_html=True)
|
| 115 |
+
|
| 116 |
+
@st.cache_resource
|
| 117 |
+
def load_language_model():
|
| 118 |
+
"""Load the language detection model"""
|
| 119 |
+
model_name = "papluca/xlm-roberta-base-language-detection"
|
| 120 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 121 |
+
model = AutoModelForSequenceClassification.from_pretrained(model_name)
|
| 122 |
+
return tokenizer, model
|
| 123 |
+
|
| 124 |
+
@st.cache_resource
|
| 125 |
+
def load_spam_model():
|
| 126 |
+
"""Load the fine-tuned BERT spam detection model"""
|
| 127 |
+
model_path = "chjivan/final"
|
| 128 |
+
tokenizer = BertTokenizerFast.from_pretrained(model_path)
|
| 129 |
+
model = BertForSequenceClassification.from_pretrained(model_path)
|
| 130 |
+
return tokenizer, model
|
| 131 |
+
|
| 132 |
+
def detect_language(text, tokenizer, model):
|
| 133 |
+
"""Detect the language of the input text"""
|
| 134 |
+
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
|
| 135 |
+
with torch.no_grad():
|
| 136 |
+
outputs = model(**inputs)
|
| 137 |
+
|
| 138 |
+
# Get predictions and convert to probabilities
|
| 139 |
+
logits = outputs.logits
|
| 140 |
+
probabilities = torch.softmax(logits, dim=1)[0]
|
| 141 |
+
|
| 142 |
+
# Get the predicted language and its probability
|
| 143 |
+
predicted_class_id = torch.argmax(probabilities).item()
|
| 144 |
+
predicted_language = model.config.id2label[predicted_class_id]
|
| 145 |
+
confidence = probabilities[predicted_class_id].item()
|
| 146 |
+
|
| 147 |
+
# Get top 3 languages with their probabilities
|
| 148 |
+
top_3_indices = torch.topk(probabilities, 3).indices.tolist()
|
| 149 |
+
top_3_probs = torch.topk(probabilities, 3).values.tolist()
|
| 150 |
+
top_3_langs = [(model.config.id2label[idx], prob) for idx, prob in zip(top_3_indices, top_3_probs)]
|
| 151 |
+
|
| 152 |
+
return predicted_language, confidence, top_3_langs
|
| 153 |
+
|
| 154 |
+
def classify_spam(text, tokenizer, model):
|
| 155 |
+
"""Classify the input text as spam or ham"""
|
| 156 |
+
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=128)
|
| 157 |
+
with torch.no_grad():
|
| 158 |
+
outputs = model(**inputs)
|
| 159 |
+
|
| 160 |
+
# Get predictions and convert to probabilities
|
| 161 |
+
logits = outputs.logits
|
| 162 |
+
probabilities = torch.softmax(logits, dim=1)[0]
|
| 163 |
+
|
| 164 |
+
# Get the predicted class and its probability (0: ham, 1: spam)
|
| 165 |
+
predicted_class_id = torch.argmax(probabilities).item()
|
| 166 |
+
confidence = probabilities[predicted_class_id].item()
|
| 167 |
+
|
| 168 |
+
is_spam = predicted_class_id == 1
|
| 169 |
+
return is_spam, confidence
|
| 170 |
+
|
| 171 |
+
# Generate and cache logo
|
| 172 |
+
logo_base64 = create_logo()
|
| 173 |
+
logo_html = f'<img src="data:image/png;base64,{logo_base64}" style="height:150px;">'
|
| 174 |
+
|
| 175 |
+
# Load both models
|
| 176 |
+
with st.spinner("Loading models... This may take a moment."):
|
| 177 |
+
lang_tokenizer, lang_model = load_language_model()
|
| 178 |
+
spam_tokenizer, spam_model = load_spam_model()
|
| 179 |
+
|
| 180 |
+
# App Header with logo
|
| 181 |
+
col1, col2 = st.columns([1, 5])
|
| 182 |
+
with col1:
|
| 183 |
+
st.markdown(logo_html, unsafe_allow_html=True)
|
| 184 |
+
with col2:
|
| 185 |
+
st.markdown('<h1 class="main-header">SMS Spam Guard</h1>', unsafe_allow_html=True)
|
| 186 |
+
st.markdown('<p class="sub-header">智能短信垃圾过滤助手 by SafeTalk Communications Ltd.</p>', unsafe_allow_html=True)
|
| 187 |
+
|
| 188 |
+
# Sidebar
|
| 189 |
+
with st.sidebar:
|
| 190 |
+
st.markdown(logo_html, unsafe_allow_html=True)
|
| 191 |
+
st.markdown("### About SafeTalk")
|
| 192 |
+
st.markdown("SafeTalk Communications Ltd. provides intelligent communication security solutions to protect users from spam and fraudulent messages.")
|
| 193 |
+
st.markdown("#### Our Technology")
|
| 194 |
+
st.markdown("- ✅ Advanced AI-powered spam detection")
|
| 195 |
+
st.markdown("- 🌐 Multi-language support")
|
| 196 |
+
st.markdown("- 🔒 Secure and private processing")
|
| 197 |
+
st.markdown("- ⚡ Real-time analysis")
|
| 198 |
+
|
| 199 |
+
st.markdown("---")
|
| 200 |
+
st.markdown("### Sample Messages")
|
| 201 |
+
|
| 202 |
+
if st.button("Sample Spam (English)"):
|
| 203 |
+
st.session_state.sms_input = "URGENT: You have won a $1,000 Walmart gift card. Go to http://bit.ly/claim-prize to claim now before it expires!"
|
| 204 |
+
|
| 205 |
+
if st.button("Sample Legitimate (English)"):
|
| 206 |
+
st.session_state.sms_input = "Your Amazon package will be delivered today. Thanks for ordering from Amazon!"
|
| 207 |
+
|
| 208 |
+
if st.button("Sample Message (French)"):
|
| 209 |
+
st.session_state.sms_input = "Bonjour! Votre réservation pour le restaurant est confirmée pour ce soir à 20h. À bientôt!"
|
| 210 |
+
|
| 211 |
+
if st.button("Sample Message (Spanish)"):
|
| 212 |
+
st.session_state.sms_input = "Hola, tu cita médica está programada para mañana a las 10:00. Por favor llega 15 minutos antes."
|
| 213 |
+
|
| 214 |
+
# Main Content
|
| 215 |
+
st.markdown('<div class="highlight">', unsafe_allow_html=True)
|
| 216 |
+
# Input form
|
| 217 |
+
sms_input = st.text_area(
|
| 218 |
+
"Enter the SMS message to analyze:",
|
| 219 |
+
value=st.session_state.get("sms_input", ""),
|
| 220 |
+
height=100,
|
| 221 |
+
key="sms_input",
|
| 222 |
+
help="Enter the SMS message you want to analyze for spam"
|
| 223 |
+
)
|
| 224 |
+
|
| 225 |
+
analyze_button = st.button("📱 Analyze Message", use_container_width=True)
|
| 226 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
| 227 |
+
|
| 228 |
+
# Process input and display results
|
| 229 |
+
if analyze_button and sms_input:
|
| 230 |
+
with st.spinner("Analyzing message..."):
|
| 231 |
+
# Step 1: Language Detection
|
| 232 |
+
lang_start_time = time.time()
|
| 233 |
+
lang_code, lang_confidence, top_langs = detect_language(sms_input, lang_tokenizer, lang_model)
|
| 234 |
+
lang_time = time.time() - lang_start_time
|
| 235 |
+
|
| 236 |
+
# Create mapping for full language names
|
| 237 |
+
lang_names = {
|
| 238 |
+
"ar": "Arabic",
|
| 239 |
+
"bg": "Bulgarian",
|
| 240 |
+
"de": "German",
|
| 241 |
+
"el": "Greek",
|
| 242 |
+
"en": "English",
|
| 243 |
+
"es": "Spanish",
|
| 244 |
+
"fr": "French",
|
| 245 |
+
"hi": "Hindi",
|
| 246 |
+
"it": "Italian",
|
| 247 |
+
"ja": "Japanese",
|
| 248 |
+
"nl": "Dutch",
|
| 249 |
+
"pl": "Polish",
|
| 250 |
+
"pt": "Portuguese",
|
| 251 |
+
"ru": "Russian",
|
| 252 |
+
"sw": "Swahili",
|
| 253 |
+
"th": "Thai",
|
| 254 |
+
"tr": "Turkish",
|
| 255 |
+
"ur": "Urdu",
|
| 256 |
+
"vi": "Vietnamese",
|
| 257 |
+
"zh": "Chinese"
|
| 258 |
+
}
|
| 259 |
+
|
| 260 |
+
lang_name = lang_names.get(lang_code, lang_code)
|
| 261 |
+
|
| 262 |
+
# Step 2: Spam Classification
|
| 263 |
+
spam_start_time = time.time()
|
| 264 |
+
is_spam, spam_confidence = classify_spam(sms_input, spam_tokenizer, spam_model)
|
| 265 |
+
spam_time = time.time() - spam_start_time
|
| 266 |
+
|
| 267 |
+
# Display Language Detection Results
|
| 268 |
+
st.markdown("### Analysis Results")
|
| 269 |
+
|
| 270 |
+
col1, col2 = st.columns(2)
|
| 271 |
+
|
| 272 |
+
with col1:
|
| 273 |
+
st.markdown("#### 📊 Language Detection")
|
| 274 |
+
st.markdown(f'<div class="result-card">', unsafe_allow_html=True)
|
| 275 |
+
st.markdown(f'<span class="language-tag">{lang_name}</span> Detected with {lang_confidence:.1%} confidence', unsafe_allow_html=True)
|
| 276 |
+
|
| 277 |
+
# Display top 3 languages
|
| 278 |
+
st.markdown("##### Top language probabilities:")
|
| 279 |
+
for lang_code, prob in top_langs:
|
| 280 |
+
lang_full = lang_names.get(lang_code, lang_code)
|
| 281 |
+
st.markdown(f"- {lang_full}: {prob:.1%}")
|
| 282 |
+
|
| 283 |
+
st.markdown(f"⏱️ Processing time: {lang_time:.3f} seconds")
|
| 284 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
| 285 |
+
|
| 286 |
+
with col2:
|
| 287 |
+
st.markdown("#### 🔍 Spam Detection")
|
| 288 |
+
|
| 289 |
+
if is_spam:
|
| 290 |
+
st.markdown(f'<div class="result-card spam-alert">', unsafe_allow_html=True)
|
| 291 |
+
st.markdown(f"⚠️ **SPAM DETECTED** with {spam_confidence:.1%} confidence")
|
| 292 |
+
st.markdown("This message appears to be spam and potentially harmful.")
|
| 293 |
+
else:
|
| 294 |
+
st.markdown(f'<div class="result-card ham-alert">', unsafe_allow_html=True)
|
| 295 |
+
st.markdown(f"✅ **LEGITIMATE MESSAGE** with {spam_confidence:.1%} confidence")
|
| 296 |
+
st.markdown("This message appears to be legitimate.")
|
| 297 |
+
|
| 298 |
+
st.markdown(f"⏱️ Processing time: {spam_time:.3f} seconds")
|
| 299 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
| 300 |
+
|
| 301 |
+
# Summary and Recommendations
|
| 302 |
+
st.markdown("### 📋 Summary & Recommendations")
|
| 303 |
+
if is_spam:
|
| 304 |
+
st.warning("📵 **Recommended Action**: This message should be blocked or moved to spam folder.")
|
| 305 |
+
st.markdown("""
|
| 306 |
+
**Why this is likely spam:**
|
| 307 |
+
- Contains suspicious language patterns
|
| 308 |
+
- May include urgent calls to action
|
| 309 |
+
- Could contain unsolicited offers
|
| 310 |
+
""")
|
| 311 |
+
else:
|
| 312 |
+
st.success("✅ **Recommended Action**: This message can be delivered to the inbox.")
|
| 313 |
+
|
| 314 |
+
# Chart for visualization
|
| 315 |
+
st.markdown("### 📈 Confidence Visualization")
|
| 316 |
+
chart_data = pd.DataFrame({
|
| 317 |
+
'Task': ['Language Detection', 'Spam Classification'],
|
| 318 |
+
'Confidence': [lang_confidence, spam_confidence if is_spam else 1-spam_confidence]
|
| 319 |
+
})
|
| 320 |
+
st.bar_chart(chart_data.set_index('Task'))
|
| 321 |
+
|
| 322 |
+
# Footer
|
| 323 |
+
st.markdown('<div class="footer">', unsafe_allow_html=True)
|
| 324 |
+
st.markdown("© 2023 SafeTalk Communications Ltd. | www.safetalk.com")
|
| 325 |
+
st.markdown("SMS Spam Guard is an intelligent message filtering solution to protect users from unwanted communications.")
|
| 326 |
+
st.markdown('</div>', unsafe_allow_html=True)
|
requirements.txt
CHANGED
|
@@ -1,3 +1,6 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
streamlit==1.32.0
|
| 2 |
+
torch==2.1.0
|
| 3 |
+
transformers==4.38.0
|
| 4 |
+
pandas==2.2.0
|
| 5 |
+
numpy==1.26.0
|
| 6 |
+
safetensors==0.4.5
|