import streamlit as st
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import numpy as np
import os
st.set_page_config(
page_title="TruthLens โ Fake News Detector",
page_icon="๐",
layout="wide",
initial_sidebar_state="collapsed"
)
st.markdown("""
""", unsafe_allow_html=True)
# โโ Load model โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
@st.cache_resource
def load_model():
from transformers import pipeline
# Using a well-trained public fake news detection model
classifier = pipeline(
"text-classification",
model="hamzab/roberta-fake-news-classification",
device=-1
)
return classifier
classifier = load_model()
# โโ Predict โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
def predict(text):
result = classifier(text, truncation=True, max_length=512)[0]
label_raw = result['label'].upper()
score = result['score']
if 'FAKE' in label_raw or label_raw == 'LABEL_0':
label = "FAKE"
fake_prob = round(score * 100, 1)
real_prob = round((1 - score) * 100, 1)
else:
label = "REAL"
real_prob = round(score * 100, 1)
fake_prob = round((1 - score) * 100, 1)
confidence = round(score * 100, 1)
risk_score = int(fake_prob)
return label, confidence, fake_prob, real_prob, risk_score
return label, confidence, round(fake_prob*100,1), round(real_prob*100,1), risk_score
def get_signals(text):
lower = text.lower()
signals = []
cap = sum(1 for c in text if c.isupper()) / max(len(text),1)
signals.append(("Capitalization", f"{cap*100:.0f}%", "HIGH" if cap>0.2 else "LOW"))
excl = text.count('!')
signals.append(("Exclamation marks", str(excl), "HIGH" if excl>=3 else ("MED" if excl else "LOW")))
sens = ['shocking','breaking','secret','hidden','suppressed','miracle','exposed','bombshell']
found = [w for w in sens if w in lower]
signals.append(("Sensational words", str(len(found)), "HIGH" if len(found)>=2 else ("MED" if found else "LOW")))
factual = ['according to','study','research','published','reported','announced']
fc = [w for w in factual if w in lower]
signals.append(("Factual language", f"{len(fc)} found", "LOW" if len(fc)>=2 else "HIGH"))
return signals
# โโ Samples โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
SAMPLES = {
"fake1": "SHOCKING: Scientists CONFIRM that drinking lemon juice cures ALL cancers overnight! Big Pharma desperately hiding this miracle cure! Share before it gets deleted!!!",
"fake2": "BREAKING: The moon landing was STAGED in Hollywood! Leaked NASA documents CONFIRM what conspiracy theorists have known. Government LYING to us for decades!!!",
"real1": "The Intergovernmental Panel on Climate Change released its assessment report indicating global temperatures have risen 1.1 degrees Celsius above pre-industrial levels, compiled by 230 scientists.",
"real2": "Apple reported quarterly revenue of $89.5 billion, with its services segment growing 16 percent year-over-year according to the company's official earnings call."
}
# โโ UI โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
# Header
st.markdown("""
v1.0 ยท PORTFOLIO PROJECT
""", unsafe_allow_html=True)
# Hero
st.markdown("""
โ AI-Powered Detection
DETECT
FAKE / REAL
NEWS
Fine-tuned DistilBERT model trained on 70,000+ articles.
Paste any headline or article to instantly analyze it.
""", unsafe_allow_html=True)
# Stats
st.markdown("""
""", unsafe_allow_html=True)
# Sample buttons
st.markdown('Try a sample โ
', unsafe_allow_html=True)
c1, c2, c3, c4 = st.columns(4)
with c1:
st.markdown('', unsafe_allow_html=True)
if st.button("๐จ Fake Sample 1"): st.session_state.text = SAMPLES["fake1"]
st.markdown('
', unsafe_allow_html=True)
with c2:
st.markdown('', unsafe_allow_html=True)
if st.button("๐จ Fake Sample 2"): st.session_state.text = SAMPLES["fake2"]
st.markdown('
', unsafe_allow_html=True)
with c3:
st.markdown('', unsafe_allow_html=True)
if st.button("โ
Real Sample 1"): st.session_state.text = SAMPLES["real1"]
st.markdown('
', unsafe_allow_html=True)
with c4:
st.markdown('', unsafe_allow_html=True)
if st.button("โ
Real Sample 2"): st.session_state.text = SAMPLES["real2"]
st.markdown('
', unsafe_allow_html=True)
st.markdown("", unsafe_allow_html=True)
# Input
text_input = st.text_area(
"input",
value=st.session_state.get('text', ''),
height=160,
placeholder="Paste any news article, headline, or suspicious text here..."
)
char_color = "rgba(255,200,0,0.7)" if len(text_input) > 0 else "rgba(255,255,255,0.2)"
st.markdown(f"{len(text_input)} CHARACTERS
", unsafe_allow_html=True)
st.markdown("", unsafe_allow_html=True)
col_btn, _ = st.columns([1, 2])
with col_btn:
st.markdown('', unsafe_allow_html=True)
analyze = st.button("โก ANALYZE NOW", use_container_width=True)
st.markdown('
', unsafe_allow_html=True)
# โโ Result โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
if analyze:
if not text_input.strip():
st.markdown("""
โ Please enter some text to analyze.
""", unsafe_allow_html=True)
else:
with st.spinner("Analyzing with DistilBERT..."):
label, confidence, fake_prob, real_prob, risk_score = predict(text_input)
signals = get_signals(text_input)
if label == "FAKE":
vc, icon, color = "verdict-fake", "โ ๏ธ", "#ff1e3c"
verdict_text = "LIKELY FAKE"
desc = "High probability of misinformation detected"
elif label == "REAL":
vc, icon, color = "verdict-real", "โ
", "#00dc82"
verdict_text = "LIKELY REAL"
desc = "Low misinformation indicators detected"
else:
vc, icon, color = "verdict-uncertain", "๐", "#ffb400"
verdict_text = "UNCERTAIN"
desc = "Mixed signals โ verify with trusted sources"
st.markdown(f"""
{icon}
{verdict_text}
{desc}
RISK LEVEL{risk_score}/100
""", unsafe_allow_html=True)
# Metrics
m1, m2, m3 = st.columns(3)
with m1:
st.markdown(f'{confidence:.1f}%
Confidence
{fake_prob:.1f}%
Fake Probability
{real_prob:.1f}%
Real Probability
{name}: {val}{badge}
'
st.markdown(f"""
Signal Breakdown
{sig_rows}
""", unsafe_allow_html=True)
# Footer
st.markdown("""
""", unsafe_allow_html=True)