VISVESVARAYA TECHNOLOGICAL UNIVERSITY 1
JNANASANGAMA, BELGAVI
CAUVERY INSTITUTE OF TECHNOLOGY
MANDYA
“NEWS LENS: AI-POWERED PERSONALIZED NEWS
AGGREGATOR”
PRESENTED BY UNDER THE GUIDANCE
1. VIKAS KUMAR G (4CA22CS116) OF
2. VARSHITH Y M (4CA22CS114) [Link] K P
3. VISWAS P M (4CA22CS119) (Assistant Professor)
4. NIKHIL GOWDA T N (4CA22CS057)
2
NEWS LENS:
AI-POWERED
PERSONALIZES
NEWS
AGGREGATOR
INTRODUCTION
3
• The rapid growth of digital media has made
news accessible to everyone, but it has also
increased the spread of misinformation across
online platforms.
• People often encounter misleading or false
content, which creates confusion and affects
public opinion. To address this challenge, this
project introduces News Lens, an AI-powered
system that analyzes full-length news articles,
verifies facts in real time, and delivers
personalized, trustworthy information.
4
PROBLEM STATEMENT
Digital platforms rapidly spread fake and
misleading news. Existing systems lack deep
contextual analysis, real-time fact checking, and
personalization, resulting in unreliable predictions.
A more accurate and transparent fake news
detection system is required.
EXISTING SYSTEM
• Older models check only basic text
features.
• Deep learning models understand better
but still miss full context.
• Some BERT-based systems check only
headlines, not full articles.
• Browser extensions warn about unsafe
websites but don’t verify news content.
• Manual fact-checking sites are accurate
but slow and require human effort.
6
PROPOSED SYSTEM
• The system reads the full article and uses
BERT to understand it.
• It checks the news in real time on trusted
websites.
• A Hybrid Decision Engine combines AI and
fact-checking for better accuracy.
• It also gives personalized news
recommendations.
• Overall, it is more accurate and user-friendly
than traditional systems.
7
OBJECTIVES
• The system reads the full article and uses
BERT to understand its meaning.
• It verifies the information by checking
trusted online sources in real time.
• It improves accuracy by combining
BERT results with fact-checking data.
• It gives personalized, reliable news
recommendations based on user interests.
8
SYSTEM SPECIFICATIONS
HARDWARE
• i3/i5 Processor
• 4–8 GB RAM
• 20+ GB Storage
• Stable Internet Connection
SOFTWARE
• Python 3.8+
• Streamlit
• Transformers (BERT)
• PyTorch
• BeautifulSoup
• GNews API
LITERATURE SURVEY 9
Pap
er Title Authors Year Methodology Merits Limitations
No.
asily
Simple, fast,
Rule-based bypassed by
helps identify
Fake News linguistic features, paraphrasing,
misleading
1 Detection on Shu et al. 2017 user-behavior shallow
patterns on
Social Media patterns, understanding
social
propagation analysis , not context-
platforms
aware
Better Fails to
Fake News
TF-IDF feature accuracy than capture deep
Classification
extraction + Logistic rule-based semantic
2 using TF-IDF + Ahmed et al. 2018
Regression and systems; meaning;
Machine
SVM classifiers efficient for high false
Learning
large datasets positives
Literature Survey 10
No external fact-
FakeBERT – Deep
BERT-based deep High accuracy, checking; cannot
Learning with
3 Kaliyar et al. 2020 learning for semantic strong contextual validate real-
BERT for Semantic
text understanding. understanding world
Understanding
correctness.
More reliable; Slow when
Multi-Source Fact- Cross-verification using credibility sources
4 Checking Hasan & Ng 2022 multiple online sources increased using unavailable;
Framework and search engines. trustworthy depends heavily
publishers. on API access.
ML classification Balanced Limited by
Hybrid ML +
combined with external prediction; better availability of
5 External Knowledge Pramanick et al. 2021
knowledge-base reasoning and verified online
Fact-Checking
credibility scores. transparency. sources
MODULES
11
1. News Collection & Article Extraction Module
Fetches latest news through API
Extracts full article text using web scraping
2. BERT-Based Classification Module
Analyzes complete article
Predicts whether the news is Real or Fake
3. Real-Time Fact Verification Module
Searches trusted sources online
Verifies authenticity of the news
4. Hybrid Decision & Recommendation Module
Combines BERT + fact-check results for final output
Provides personalized and verified news recommendations
12
System Architecture
13
14
IMPLEMENTATION STEPS
1. Data Collection & Article Extraction
Fetch latest news using GNews API
Extract full article text through web scraping
2. Preprocessing & BERT Integration
Clean and prepare article text
Feed the processed text into the fine-tuned BERT model
3. Fake News Classification
BERT analyzes full content
Generates prediction: Real / Fake
15
IMPLEMENTATION STEPS
1. Data Collection & Article Extraction 5. Hybrid Decision Generation
Fetch latest news using GNews API •Combine BERT output + fact-check results
Extract full article text through web scraping •Produce final verdict: Real / Fake / Unverified
2. Preprocessing & BERT Integration 6. User Interface Development
Clean and prepare article text •Build Streamlit dashboard
Feed the processed text into the fine-tuned BERT •Display article, prediction results, and verification details
model
7. Personalized Recommendation
3. Fake News Classification •Track user reading patterns
BERT analyzes full content •Suggest verified and relevant news articles
Generates prediction: Real / Fake
4. Real-Time Fact Verification
•Search trusted sources (BBC, Reuters, NDTV etc.)
•Match and validate similar news online
16
SYSTEM TESTING
1. Performance Testing
• Ensure fast article extraction, quick predictions, and real-
time fact checking.
• Check response time under multiple news requests.
2. Accuracy Testing
• Compare model predictions with labeled datasets.
• Measure precision, recall, and overall accuracy of BERT-
based classification.
3. UI/UX Testing
• Test dashboard navigation, clarity, and user interaction.
• Ensure smooth display of results and recommendations.
FUTURE
ENHANCEMENTS
1. Multilingual Fake News Detection
2. Deepfake Image/Video Analysis
3. Larger & Locally Fine-Tuned
Transformer Models
4. Browser Extension Version
5. Database-Driven User Profiles
6. ML + Knowledge Graph
Integration
SNAPSHOTS
Fig: Home Page / Dashboard Fig: Category Selection Screen
19
Fig: News Fetch Result Page Fig: Full Article
Extraction Screen
20
Fig: Fact-Checking Result Screen Fig: BERT Prediction Output Screen
Fig:Final Hybrid Decision Output Fig: Recommendation System Page
[Link]
APPLICATIONS 22
1. News Verification Platforms
2. Social Media Monitoring
3. Journalism & Media Houses
4. Government & Public Safety
5. Education & Research
6. Personalized News Apps
CONCLUSION 23
This project helps detect fake news by using
AI to read the full article and check the
information with trusted news sources. The
BERT model analyzes the content, and real-
time fact-checking confirms its accuracy. By
combining both methods, the system gives
more accurate and reliable results than
traditional tools. It also recommends verified
news to users, helping reduce
misinformation and making online news
safer and more trustworthy.
