AI-POWERED NEWS
RECOMMENDER BOT
ENHANCING USER EXPERIENCE AND
REDUCING INFORMATIONOVERLOAD
TEAM:-
HARSH VARDHAN SINGH, 2K23CSUN01210
KRISH SAINI, 2K23CSUN01218
SHIVAM AGGARWAL, 2K23CSUN01230
INTRODUCTION
In today's digital age, information overload is a growing challenge.
With thousands of news articles published every minute, it becomes
increasingly difficult for users to find relevant content that matches
their interests.
Our project — AI-Powered News Recommender Bot — aims to solve
this problem by leveraging Artificial Intelligence and Machine
Learning techniques to deliver personalized news recommendations.
By analyzing user behavior, preferences, and real-time data, our bot
intelligently curates content that is both relevant and timely.
This not only enhances user experience but also ensures that
important and interesting stories never go unnoticed.
01
WHY NEWS RECOMMENDER BOT?
Ease of Access: Bots can be integrated into platforms
users already use (like WhatsApp, Telegram, or
websites), removing the need to download or install
anything.
Faster Interaction: Bots provide a conversational and
intuitive interface, making the user experience more
natural and engaging.
Lightweight and Efficient: Bots consume fewer
resources compared to apps and are easier to maintain
and update.
Wider Reach: Bots are more platform-agnostic, reaching
users across devices without compatibility issues.
Our goal is to deliver the news people care about —
quickly, intelligently, and effortlessly.
02
Other Apps [Link]
Why
Choose
Us?
API USED (NewsApi)
1. Why News API?
✅ Comprehensive Coverage – Fetches news from
Working of Api
50,000+ sources (BBC, CNN, Times of India, etc.)
✅ Real-Time Updates – Articles are refreshed every
10-15 minutes
✅ Geographic Filtering – Supports region-specific
queries (e.g., q=India)
✅ Free Tier Available – 500 requests/day (sufficient
for small-scale bots) 2. Rate Limits & Optimization
Free Tier Limits:
500 requests/day
Response of API
1 request/second
Optimization Tips:
Cache responses for 10 mins (avoid
redundant calls)
Use pageSize=50 to minimize API hits
Implement error handling for 429 Too Many
03
Requests
-TECHNOLOGY USED-
Text Processing (TF-IDF Vectorizer)
The TfidfVectorizer converts raw news text into numerical data by weighing
word importance. It filters out common English stopwords ("the", "and") to
focus on meaningful keywords. By limiting to 1,000 features (max_features),
it reduces noise while preserving key patterns. This structured data enables
the ML model to analyze article similarities effectively.
Clustering (K-Means)
K-Means groups articles by similarity using their TF-IDF vectors,
automatically detecting topics like "Sports" or "Politics". The dynamic cluster
count (min(10, len(articles)//2)) ensures balanced grouping without
overfitting. This eliminates duplicate news, providing users with diverse
stories. Clustering is efficient, processing 50+ articles in seconds.
Summarization (LSA Summarizer)
The LsaSummarizer extracts key sentences using Latent Semantic Analysis,
condensing articles into 3-sentence summaries. It identifies core themes
while ignoring redundant details, mimicking human skimming. For short
texts, it falls back to snippets to ensure readability. This delivers crisp,
actionable news without overwhelming users.
04
METHODOLOGY
1. Data Collection (NewsAPI Integration)
Input: User queries (e.g., "Cricket news India")
Process:
Fetches real-time articles via NewsAPI with filters (language=en, sortBy=publishedAt).
Validates articles (removes empty titles/URLs).
Output: Raw JSON news data (title, content, source, URL).
2. Machine Learning Processing
(A) Text Vectorization (TF-IDF)
Converts text into numerical vectors.
Filters stopwords (e.g., "the," "and") and limits to top 1,000 keywords.
Why? Enables semantic analysis for clustering.
(B) Clustering (K-Means)
Groups similar articles using TF-IDF vectors.
Dynamic clusters: min(10, len(articles)//2) to avoid over/under-clustering.
Why? Eliminates duplicate news (e.g., 5 articles on the same event).
(C) Summarization (LSA)
Extracts top 3 sentences per article using Latent Semantic Analysis.
Falls back to snippets for short texts.
Why? Delivers concise, scannable summaries.
3. Bot Response Generation
Formats clustered/summarized news with Markdown (bold titles, source links).
Sends responses via python-telegram-bot with rate limiting (sleep(1)).
05
VISUAL REPRESENTATION
07
ARCHITECTURE STRUCTURE
06
UI of the app
UI of the app
[Link] 4 - Quality Education: Provides easy
access to reliable and personalized news.
[Link] 16 - Peace, Justice, and Strong
Institutions: Helps combat misinformation
and promotes informed decision-making.
09
CONCLUSION
We’ve successfully built an AI-powered Telegram bot
that:
✅ Fetches real-time news from multiple sources
✅ Uses NLP clustering to group similar articles
✅ Generates concise summaries for quick reading
✅ Allows custom topic searches (no fixed menus)
Future Plan
📱 App Version
- Same smart news filtering
- +Offline reading, notifications, and
Why Start with a Bot? user profiles
- Faster testing of core AI features
- No app store approvals needed
- Easy to share and demo
10
THANK YOU
11
