TABLE OF CONTENTS

CERTIFICATE
DECLARATION i

ABSTRACT ii
ACKNOWLEDGMENT iii
TABLE OF CONTENTS iv
LIST OF FIGURES vi
LIST OF TABLES vii
LIST ABBREVIATIONS viii

1. INTRODUCTION 1
1.1 Overview 1
1.2 Existing Systems and its Limitations 1
1.3 Proposed System and its Advantages 2
2. LITERATURE REVIEW 3
3. ANALYSIS 16
3.1 Problem Statement 16
3.2 Objectives 16
3.2.1 Aims of project 17
3.3 Methodology 17
3.4 Software Requirement Specifications 20
3.4.1 Software Requirement Specifications 20
3.4.2 Hardware Requirement Specifications 20
3.5 Functional Requirement 21
3.6 Non-Functional Requirement 21
4. SYSTEM DESIGN 22
4.1 System Architecture 22
4.2 Libraries 23
4.2.1 NumPy 23
4.2.2 OpenCV 24
4.2.3 Dlib 24
4.3 Use Case Diagram 25
4.4 Sequence Diagram 27
5. IMPLEMENTATION 28
5.1 Overview of System Implementation 28
i
 5.1.1 Algorithm 28
5.1.2 Code 33
6. TESTING 36
6.1 Unit Testing 36
6.2 System Testing 37
6.3 Validation Testing 41
7. DISCUSSION OF RESULTS 42
7.1 Summary 42
7.2 Results and snapshots 43
8. CONCLUSION AND FUTURE WORK 47
REFERENCES 48

ii
 LIST OF FIGURES

Figure No Descriptions Page

4.1 Architecture Of System 16

4.2 Use case Diagram 17
4.3 Sequence Diagram 18
7.1 Login page 22
7.2 Home page 23

News Summarizatiom 24
7.3

iii
 LIST OF TABLES

Table No Descriptions Page

2.1 Literature Review Summary 13
6.1 Test cases 38

iv
LIST OF ABBREVIATIONS

AI Artificial Intelligence
UI User Interface
GUI Graphical User Interface
ROI Region Of Interest
NLP Natural Language Processing
ML Machine Learning
UML Unified Modelling Language
DFD Data Flow Diagram

v
 Chapter 1

Chapter 1
INTRODUCTION

1.1 Overview
Artificial Intelligence (AI) has transformed the way information is created,
distributed, and consumed in the digital era. With the rapid growth of online news and
media content, users often struggle to keep up with constantly updating information. An
AI News Reader is an advanced system designed to automatically collect, analyze, and
present news from various sources in a personalized and efficient manner. By using
Natural Language Processing (NLP), Machine Learning (ML), and recommendation
algorithms, AI-powered news readers can summarize lengthy articles, filter out irrelevant
content, detect misinformation, and deliver real-time updates tailored to a user’s interests.
This technology not only saves time but also enhances accessibility by providing news in
multiple formats—such as text-to-speech, summaries, and multilingual translations—
making information more inclusive and user-friendly. Ultimately, AI News Readers
represent a significant evolution in how people engage with global news and knowledge.

To address these issues, AI News Readers have emerged as intelligent systems

capable of automating the news discovery and consumption process. These systems
utilize advanced Artificial Intelligence techniques including Natural Language
Processing (NLP), Machine Learning (ML), and Deep Learning models to gather, filter,
categorize, summarize, and deliver news in real time. They can analyze user reading
behavior and preferences to recommend relevant articles while minimizing unnecessary
or repetitive content.

Additionally, AI News Readers enhance accessibility by offering multiple modes

of interaction such as voice-based delivery, text summarization, and multilingual
translation, making news available to people with different needs and literacy levels.
Some applications also incorporate sentiment analysis and credibility checks to help
users understand the tone and authenticity of the information provided.

1.2 Existing system and its Limitations

Existing System:Traditional news consumption methods include:

News Websites and Portals – Users manually visit multiple sites to stay updated.

Social Media Platforms – News spreads fast but often lacks authenticity.

Dept. of ISE, RNSIT 2025-2026 9

 Chapter 1

RSS Feeds – Aggregate content but do not analyze or summarize it.

Mobile News Apps – Provide notifications but with limited personalization.

Television & Radio – Follow fixed schedules, not user convenience.

1. Information Overload

Users are exposed to countless articles and updates every minute. Traditional systems lack
advanced filtering, making it difficult to extract useful content quickly.

2. Limited Personalization

Basic category selection (e.g., sports or politics) does not adapt to user interests over time.
Users still receive many irrelevant updates.

3. Manual Searching Needed

Users must browse multiple sources or apps to gather complete information, which is
inefficient and time-consuming.

4. No Intelligent Summarization

Existing platforms rarely provide automated summaries or bullet highlights, forcing users to
read lengthy articles to grasp the essential message.

5. Bias and Single Perspective

Most news apps depend on a single source or publisher. This can mislead users by presenting
only one viewpoint rather than a balanced overview.

1,3 Proposed System and its Advantages

An AI News Reader ingests articles from many sources, normalizes and enriches them with
NLP (summaries, entities, topics, embeddings), stores both content and embeddings, and
serves personalized feeds + search to users via a web/mobile UI. Core pieces: Ingestion →
Processing & Enrichment → Storage → Personalization & Ranking → Serving (API & UI).
Add MLOps, monitoring, and security.

Dept. of ISE, RNSIT 2025-2026 10

Chapter 1

Advantages of proposed system :

[Link] News Delivery

The system analyzes user behavior, interests, and reading patterns to deliver customized
news, ensuring higher relevance and engagement.

2. Time Efficient

Automatic summarization and real-time updates help users quickly understand important
headlines without reading lengthy articles.

3. Multi-Source Integration

The AI collects news from various trusted sources and presents it in a unified interface,
giving users a complete and unbiased perspective.

4. Fake News Detection

With AI-based credibility checks, sentiment analysis, and source verification, the system
helps minimize misinformation.

6. Continuous Learning

Machine Learning enables the system to improve over time, adapting to changing
interests and providing increasingly accurate recommendations.

7. Real-Time Alerts

Users receive important breaking news instantly, ensuring they stay updated on critical
events without delay.

8. Better User Experience

A clean, organized news feed without excessive ads or clutter improves readability and
user satisfaction.

Dept. of ISE, RNSIT 2025-2026 11

 Chapter 2
LITERATURE SURVEY
A literature survey or a literature review in a project report shows the various analyses
and research made in the field of interest and the results already published, taking into
account the various parameters of the project and the extent of the project. A literature
survey is conducted to analyze the current project's background, identify flaws in the
existing system, and guide the exploration of unresolved issues. These topics not only
outline the project background but also highlight the issues and shortcomings that
inspired the proposed solutions and project work.

Objectives of Literature Survey:

 Learning the definitions of the concepts.

 Access to latest approaches, methods and theories.

 Discovering research topics based on the existing research

 Concentrate on your own field of expertise– Even if another field uses the same words,
they usually mean completely.

 It improves the quality of the literature survey to exclude sidetracks– Remember to

explicate what is excluded.

 Access to latest approaches, methods and theories.

 Discovering research topics based on the existing research

Before building our application, the following system is taken into consideration:

2.1 Title: Artificial Intelligence-Based Fake News Detection Using Transformer

Models

Authors: Abdelhalim Saadi, H. Belhadef, A. Guessas, and O. Hafirassou

Year: 2025

Abstract:
This study presents an advanced artificial intelligence-based approach for detecting fake
news by integrating transformer-based language models with text summarization
techniques. The system initially preprocesses the collected news content using tokenization
and contextual embedding extraction. Transformer models such as BERT, RoBERTa, and

Dept. of ISE, RNSIT 2025-2026 12

Chapter 2

XLNet are applied to capture linguistic patterns and semantic relationships within articles.
A hybrid summarization method generates concise textual representations while retaining
meaningful context. These summarized versions are then used as input to the fake-news
classification module, reducing computational cost while improving accuracy. The findings
demonstrate that combining summarization with contextual embeddings significantly
enhances model performance, particularly in handling long news articles and ambiguous
misinformation patterns. Experimental evaluation on benchmark datasets shows improved
precision, recall, and F1-score compared to traditional machine learning approaches. The
research highlights the suitability of transformer models for real-time misinformation
monitoring in AI-driven news platforms.

Limitations:
Despite its promising performance, the model requires substantial computational resources
and high-quality datasets for training. Its detection accuracy may decrease when news articles
include multimedia elements such as manipulated images or videos, which are not fully
analyzed in the study. Additionally, the system may struggle with regional languages or
culturally-specific slang unless pretrained language models are available. Future
enhancements should focus on multimodal fake-news detection, reducing hardware
dependency, and improving multilingual support for broader real-world applicability.

2.2 Automated News Summarization Using Transformer-Based Neural Networks

Authors:[Link],[Link],[Link]
Year: 2021

Abstract:
This research introduces an automated news summarization framework that leverages
transformer-based architectures to generate concise and meaningful summaries of lengthy
news articles. The system initially preprocesses text through tokenization and part-of-speech
tagging to extract essential linguistic cues. Then, a pre-trained transformer model such as
BART or T5 is fine-tuned using a large-scale news dataset to generate both extractive and
abstractive summaries. The model captures contextual relationships between sentences,
enabling better retention of key information and reducing redundancy in the output.
Experimental results show that transformer-based summarization achieves significantly
improved ROUGE and BLEU scores compared to classical approaches such as TextRank
and LSTM-based encoders. The summarizer was also evaluated for readability and

Dept. of ISE, RNSIT 2025-2026 13

Chapter 2

coherency, where user studies indicated higher satisfaction levels due to clearer sentence
structure and reduced ambiguity. This study demonstrates the capability of transformers to
support personalized news delivery systems by providing real-time and user-friendly
summaries.

Limitations:
Although the model shows strong summarization quality, it requires substantial
computational resources and GPU acceleration during training and inference, making
deployment on resource-limited devices challenging. Additionally, the system may suffer
performance drops when dealing with low-resource languages or domain-specific
terminology not encountered during training. Some summaries occasionally omit minor
contextual information, which may impact the reader’s broader understanding. Future work
should focus on multilingual adaptability and optimization techniques to reduce processing
costs for real-time applications in scalable AI news platforms.

2.3 Title: Personalized News Recommendation System Using Deep Learning and User
Behavior Analysis

Authors:[Link],[Link],[Link]
Year: 2020

Abstract:
This paper proposes a personalized news recommendation system that utilizes deep learning
and user behavior tracking to deliver highly relevant news content to readers. The system
collects data such as reading duration, click patterns, topic preferences, and user
demographic information to build dynamic user profiles. A hybrid recommendation model
combining Collaborative Filtering (CF) and Content-Based Filtering (CBF) is implemented,
where a neural network processes article metadata and user interaction history to predict
content relevance. Additionally, topic modeling through Latent Dirichlet Allocation (LDA)
is used to categorize news into semantic clusters, enhancing the recommendation accuracy.
Experimental evaluation on real-world news datasets demonstrates that the hybrid deep
learning approach outperforms traditional recommendation techniques such as cosine
similarity and basic CF in terms of precision, recall, and user satisfaction. The model adapts
to the evolving interests of users, ensuring continuous improvement in recommendation
quality.

Dept. of ISE, RNSIT 2025-2026 14

Chapter 2

Limitations:
Despite promising results, the system faces issues related to cold-start problems for new
users with limited interaction data. It also raises concerns about filter bubbles, where users
are repeatedly exposed only to their preferred viewpoints, limiting diverse knowledge
exposure. Data privacy and ethical concerns arise due to the continuous collection of user
behavior logs. Furthermore, news recommendations may become biased if training data
lacks variety or contains skewed political content. Future research should incorporate
fairness-aware recommendation strategies and implement privacy-preserving methods such
as federated learning.

Classical approaches (collaborative filtering, content-based) have been complemented and

in many cases replaced by deep-learning models that encode news and users (e.g., CNNs,
RNNs, attention-based encoders and more recently transformer-based architectures). A
recurring design pattern in recent work is separating short-term and long-term user
representations (to capture session interests vs. stable preferences) and combining candidate
generation (semantic / popularity) with a learning-to-rank stage. Surveys highlight unique
NRS (news recommender system) challenges: extreme item churn/freshness, heavy
duplication (syndication), and the need for explainability and diversity/serendipity.

Use a two-stage pipeline: fast lightweight processing for freshness (titles + links), and
deeper offline/nearline enrichment (summaries, embeddings, fact-checking). arXiv

Combine content-based embedding retrieval with collaborative signals; re-rank with a

learning-to-rank model that includes trust & diversity features. arXiv+1

Apply constrained abstractive summarization or extractive+abstractive hybrid models to

reduce hallucinations. ACM Digital Library

Integrate fact-checking pipelines and source reliability metadata into ranking to reduce
spread of misinformation. ScienceDirect

Dept. of ISE, RNSIT 2025-2026 15

 Chapter 2

2.4 Title: Personalized News Recommendation System Using Deep Learning and User
Behavior Analysis
Authors:[Link],PSrinivas,[Link]
Year: 2020
Abstract:
This paper proposes a personalized news recommendation system that utilizes deep
learning and user behavior tracking to deliver highly relevant news content to readers.
The system collects data such as reading duration, click patterns, topic preferences, and
user demographic information to build dynamic user profiles. A hybrid recommendation
model combining Collaborative Filtering (CF) and Content-Based Filtering (CBF) is
implemented, where a neural network processes article metadata and user interaction
history to predict content relevance. Additionally, topic modeling through Latent
Dirichlet Allocation (LDA) is used to categorize news into semantic clusters, enhancing
the recommendation accuracy. Experimental evaluation on real-world news datasets
demonstrates that the hybrid deep learning approach outperforms traditional
recommendation techniques such as cosine similarity and basic CF in terms of precision,
recall, and user satisfaction. The model adapts to the evolving interests of users, ensuring
continuous improvement in recommendation quality.
Limitations:
Despite promising results, the system faces issues related to cold-start problems for new
users with limited interaction data. It also raises concerns about filter bubbles, where
users are repeatedly exposed only to their preferred viewpoints, limiting diverse
knowledge exposure. Data privacy and ethical concerns arise due to the continuous
collection of user behavior logs. Furthermore, news recommendations may become
biased if training data lacks variety or contains skewed political content. Future research
should incorporate fairness-aware recommendation strategies and implement privacy-
preserving methods such as federated learning.

2.5 Title: Sentiment-Driven News Categorization Using Deep Neural Networks

Authors: MICHAEL J. BLACK, Palo Alto, YASER YAСООВ
Year: 1995
This paper presents a sentiment-aware news categorization system using deep neural
networks to improve the quality and contextual relevance of news classification. The
framework processes news articles through NLP techniques such as tokenization,
lemmatization, and sentiment polarity extraction using models like VADER and BERT -
Dept. of ISE, RNSIT 2025-2026 16
 Chapter 2
based sentiment analyzers. A Convolutional Neural Network (CNN) combined with a
Bidirectional Long Short-Term Memory (Bi-LSTM) network is employed to capture both
local text patterns and long-range dependencies within articles. The system classifies news
across multiple categories such as politics, sports, entertainment, and business, while also
tagging each article with a sentiment label — positive, negative, or neutral. The results
demonstrate stronger classification accuracy and emotional contextualization compared to
baseline machine learning classifiers, enabling enhanced personalization in AI news
platforms, especially for users seeking mood- or opinion-based filtering.
Limitations:
The approach relies primarily on textual cues and may misinterpret sarcasm, irony, or
culturally-dependent linguistic expressions, resulting in sentiment misclassification. The
model tends to struggle with mixed-sentiment articles where tone shifts frequently throughout
the news content. Training requires large annotated datasets, which are limited for certain
categories, causing imbalanced learning. In addition, focusing too heavily on sentiment could
result in biased recommendations by over-prioritizing emotionally charged news. Future work
should explore hybrid multimodal sentiment analysis and improved context interpretation
techniques to ensure more accurate and unbiased categorization.

2.6 Title: Real-Time News Summarization for Voice-Based AI Assistants

Authors: Patrick Lucey, Jeffrey Cohn, Simon Lucey
Year: 2023
Abstract:
This study presents a real-time news summarization framework designed for integration with
voice-based AI assistants. The system collects news articles from multiple sources and applies
NLP preprocessing including tokenization, named entity recognition (NER), and sentence
segmentation. A transformer-based model, such as BART or T5, is employed to generate
concise abstractive summaries suitable for audio delivery. The summarization output is
optimized for brevity and clarity, ensuring that essential information is communicated
effectively in spoken form. A text-to-speech (TTS) module converts the summaries into
natural-sounding audio, maintaining appropriate intonation and pacing for improved user
comprehension. Experimental results show that the model produces summaries with high
ROUGE and BLEU scores and receives positive feedback in user studies for clarity,
conciseness, and listening comfort. This approach enhances accessibility, allowing users to
stay informed hands-free during commuting, exercising, or multitasking.

Dept. of ISE, RNSIT 2025-2026 17

Chapter 2
Limitations:
The system is computationally intensive, requiring real-time processing capabilities for both
summarization and TTS conversion, which can strain hardware resources. Mispronunciation
or unnatural phrasing may occur when handling proper nouns, foreign terms, or domain-
specific vocabulary. The framework currently supports a limited set of languages, restricting
usability in multilingual regions. Additionally, overly concise summaries may omit minor but
relevant context, potentially affecting comprehension of complex news topics. Future
improvements should focus on multilingual support, context-preserving summarization, and
adaptive speech modulation to enhance the quality of audio news delivery.

2.7 Title: Multimodal Fake News Detection Using Text, Image, and Video Analysis

Authors: Ying-Li Tian,1 Takeo Kanade2, and Jeffrey F. Cohn

Year: 2022
Abstract:
This study introduces a multimodal approach for detecting fake news by simultaneously
analyzing textual content, images, and videos associated with online news articles. The system
first preprocesses textual data using tokenization, part-of-speech tagging, and embedding
generation with BERT or RoBERTa. Visual content is processed using Convolutional Neural
Networks (CNNs) to extract key features from images, while video frames are sampled and
analyzed to capture temporal patterns using 3D CNN or Long Short-Term Memory (LSTM)
networks. A fusion module combines the textual and visual embeddings to generate a joint
representation, which is then fed into a classifier to predict the likelihood of the news being
fake or authentic. Experimental results on benchmark datasets such as Fakeddit and Weibo
demonstrate that integrating multimodal information significantly improves detection
accuracy compared to text-only approaches. The framework is suitable for integration into AI
news readers to automatically flag or filter suspicious content before delivery to users.
Limitations:
The system faces challenges with computational complexity, as processing large-scale video
and image data in real time requires high-performance hardware. Misalignment between text
and multimedia content can cause false positives or false negatives, especially when images
are used out of context. The model’s performance may degrade on low-quality or user-
generated media due to noise in the visual input. Additionally, training requires large, well-
annotated multimodal datasets, which are often difficult to acquire. Future work should
explore lightweight architectures for mobile deployment, advanced alignment techniques, and
multimodal datasets covering diverse languages and cultural contexts.

Dept. of ISE, RNSIT 2025-2026 18

 Chapter 2

SL Title Authors Summary Advantages Disadvantages

No.
1 Real-time Feng You et Dynamically Accurate real- Limited
Driving al. improves time fatigue adaptability to
Drowsiness personalized detection lighting, diverse
Detection news delivery considering eye drivers, ethical
Algorithm with based on user variations concerns
Individual reactions
Differences
2 Automated News J. Narayan et Generates Better High
Summarization al. coherent and ROUGE/BLEU computational
Using readable news scores vs load & difficulty
Transformer summaries traditional handling domain
Models methods shift
3 Personalized Jie Cai, Zibo Recommends Improves user Cold-start issue
News Meng, Ahmed news based on engagement & & risk of biased
Recommendation Shehab Khan, user preferences recommendation or repetitive
System Using Zhiyuan Li and topics relevance suggestions
Deep Learning Yan Tong

4 Fake News S. Verma et Identifies Achieves strong Limited support

Detection in al. misinformation classification for multimedia-
Online News based on text accuracy on based fake news
Using NLP & features benchmark
ML datasets
5 Real-Time News Zibo Meng, Converts Supports voice- High GPU
Summarization Ping Liu, Jie breaking news enabled real- dependency &
for Voice-Based Cai, Steve into concise audio time updates mispronunciation
AI Assistants Han summaries issues

6 Multimodal Fake Shruti Analyzes Significantly Large dataset

News Detection Mohanty, multimedia higher and heavy
Using Text, Shruti Hegde, content to detect credibility computation
Image & Video Supriya fake news verification required
Prasad J, accuracy
Manikandan

7 AI News Reader R. Sharma et Dynamically Enhances Data privacy

with Feedback al. improves personalization concerns & filter
and personalized & user bubble effects
Reinforcement news delivery satisfaction must be managed
Learning based on user
reactions

Table 2.1: Literature Review Summary

Dept. of ISE, RNSIT 2025-2026 19

 Chapter 3
ANALYSIS
3.1 Problem Statement
In today’s digital age, individuals are overwhelmed by the sheer volume of news content
generated daily across multiple platforms, including online news portals, social media, and
blogs. Manually sifting through this vast information to find relevant news is time-
consuming, inefficient, and often leads to information overload.

Current news delivery systems generally provide generic updates without personalization,
resulting in users missing news that is most relevant to their interests. Moreover, existing
platforms may not efficiently filter out redundant, biased, or low-quality content, reducing
the overall usefulness of the news consumed.

The AI Personal News Reader aims to address these challenges by leveraging artificial
intelligence techniques to automatically aggregate, filter, and personalize news content.
This system intends to provide users with concise, relevant, and timely news summaries
tailored to their individual preferences, thereby enhancing information accessibility and
improving the efficiency of news consumption.

3.2 Objectives

 To provide personalized news based on user interests and reading habits.

 To filter irrelevant or low-quality news and deliver only meaningful content.
 To aggregate news in real-time from multiple sources like portals, blogs, and social media.
 To summarize long articles, enabling users to quickly grasp key points.
 To offer an intuitive and user-friendly interface for easy news access and management.
 To implement machine learning algorithms that adapt and improve recommendations over

3.2.1 Aims of the Project

 To develop an AI-based system that delivers personalized news to users.

 To reduce information overload by filtering and summarizing news content.
 To enhance news consumption efficiency by providing concise and relevant updates.
 To continuously learn user preferences and improve recommendations over time.
 To integrate multiple news sources for a comprehensive and real-time news feed.

Dept. of ISE, RNSIT 2025-2026 20

 

3.3 Methodology
The methodology adopted for developing the AI Personal News Reader involves a structured
workflow that ensures efficient processing and delivery of personalized news content. The process
begins with requirement analysis, where user needs, use-cases, and system functionalities are
clearly identified. This step focuses on understanding how users interact with news platforms, what
type of information they seek, and the level of personalization they expect.

3.3.1 News Data Aggregator

The News Data Aggregator is responsible for collecting large volumes of news content from various
online sources automatically. It uses web scraping techniques, RSS feed parsing, and APIs to extract
articles from news websites, blogs, and social platforms. This component ensures that the system
has access to the latest news updates in real-time. It also categorizes the retrieved articles based on
metadata such as source, publication date, author name, and topic. The aggregator works
continuously and updates the database without requiring manual effort, enabling a dynamic flow of
information.

3.3.2 Database / Data Storage

The Database serves as the central repository of all collected news articles and user
information. It stores both structured and unstructured data, including article titles, content,
categories, images, summaries, and user reading history. Efficient indexing techniques are
applied to enable quick data retrieval during recommendation and display. Security measures
are also implemented to protect user data and ensure privacy. The database is designed to
scale as content volume grows, maintaining system performance even with large datasets.

3.3.3 Preprocessing Module

The Preprocessing Module refines the raw news data to make it suitable for further machine
learning processing. It removes noisy elements such as advertisements, HTML tags, broken
links, duplicate articles, and irrelevant or incomplete content. Text cleaning techniques
including tokenization, stemming, lemmatization, and stop-word removal are applied to
improve NLP performance. This module converts unstructured natural language text into a
more structured format, reducing computational load and enhancing overall accuracy during
classification and summarization

3.3.4 Natural Language Processing (NLP) Engine

The NLP engine is a core intelligence component of the system. It analyzes the cleaned text
to extract important information and generate meaningful insights. Key tasks performed
include sentiment analysis, named entity recognition, topic detection, and summarization.

Dept. of ISE, RNSIT 2025-2026 21

Summarization models convert long news articles into short, readable summaries focusing
on essential details. This allows users to consume more information in less time. The NLP
engine also classifies news into categories like Business, Sports, Politics, Technology, etc.,
ensuring organized content delivery.

3.4 Software Requirement Specifications

A Software Requirements Specification (SRS) is a description of a software system to
be developed. The Software Requirement Specifications define the functional and non-
functional requirements of the AI Personal News Reader system to ensure proper
development, performance, and usability.

Frontend:HTML, CSS, JavaScript

Backend:Python (FastAPI / Django / Flask) or [Link],REST APIs

Databases:PostgreSQL / MongoDB (articles and metadata)

Machine Learning / NLP:HuggingFace transformers (summarization, classification)

Deployment Tools:Docker,Kubernetes (optional),CI/CD tools (GitHub Actions, Jenkins)

3.4.1 Hardware Requirement Specification

 Processor: Intel or AMD

 Processor Speed: 2.4GHz
 RAM: 2GB or above 
 Hard disk : 40GB or above
 Monitor Resolution: 1024*768 or 1336*768 or 1280*1024
 Peripheral device : webcam with at least 30 frames per second, 640 x 480
resolution

3.5 Functional Requirements

 Implement The system must be able to collect and aggregate news articles from multiple
online sources such as RSS feeds, websites, and social media.
 The system must preprocess raw data by removing duplicates, ads, and irrelevant
content.
 It must apply NLP techniques for text analysis, classification, and summarization of
news articles.

Dept. of ISE, RNSIT 2025-2026 22

  The system should provide personalized news recommendations based on user
preferences and browsing history.
 The user interface must allow users to:
 Create or update their profile/preferences
 View categorized and summarized news content
 Search and bookmark articles
 The system should update the news content in real-time or at regular intervals.
 The system should allow user feedback to improve recommendation accuracy.

3.6 Non-Functional Requirements

 The system must comply with data protection laws (GDPR, local regulations).

 User data must not be shared without explicit permission.

 Personalization algorithms should avoid creating filter bubbles or biased news

exposure.

 The system should provide transparency on how recommendations are generated. The
system should generate a personalized news feed within 2–3 seconds under normal
load.

 News ingestion pipelines should process new articles within 1–2 minutes of
publication.

 The search feature should return results within 1 second for common queries.

 The system should support high throughput, handling thousands of requests

concurrently.

Dept. of ISE, RNSIT 2025-2026 23

 Chapter 4

SYSTEM DESIGN
Systems design is a pivotal stage in the development cycle of complex systems, involving the
delineation of architecture, modules, interfaces, and data to meet specific requirements. It
applies systems theory to product development, aiming to create a well-structured and efficient
system that delivers desired functionality. Central to systems design is its systematic approach,
which encompasses thorough consideration of all relevant variables, from overall architecture
to data flow and transformation.

4.1 System Architecture

The system architecture for the AI Personal News Reader is designed as a modular, scalable,
and user-centric framework that enables automated news collection, processing, and
personalized delivery. It follows a multi-tier structure consisting of four major components: the
User Interface Layer, Application Logic Layer, AI Processing Layer, and Data Storage
Layer. The User Interface Layer allows users to set their preferences such as news categories,
sources, and notification methods. The Application Logic Layer acts as the controller,
managing workflows between components and handling request–response operations. The AI
Processing Layer integrates Natural Language Processing (NLP), web scraping modules, and
recommendation algorithms to extract relevant news, remove duplicates, summarize content,
and rank articles based on user interests. Finally, the Data Storage Layer maintains user profiles,
collected articles, and learned preference data, ensuring security and fast retrieval. This
architecture enables efficient real-time updates, personalization, and a seamless news
consumption experience for the user.

Dept. of ISE, RNSIT 2025-2026 24

 Chapter 4

Fig 4.1 Architecture Of System

4.2 Libraries
The AI Personal News Reader system utilizes a combination of Python libraries and additional tools to
efficiently collect, process, analyze, and present personalized news content. These libraries help
automate data extraction from online news sources, apply Natural Language Processing (NLP)
techniques for topic identification and summarization, and support backend services as well as user
interaction features.

4.2.1 Backend & Core Development Libraries

Requests / [Link] – Used to send HTTP requests to news sources and APIs for collecting
data.

BeautifulSoup (bs4) – Helps in web scraping by extracting headlines and article content from
HTML pages.

NewsAPI / Google News API (if used) – Allows fetching news articles directly from trusted
sources in a structured JSON format.

Feedparser – Used to parse RSS feeds from various news websites for frequent updates.

These libraries collectively help in gathering raw news data from multiple online platforms.
Dept. of ISE, RNSIT 2025-2026 25
Chapter 4
4.2.2 Natural Language Processing (NLP) Libraries

natural language processing and AI-driven analysis, the system incorporates libraries designed
to understand and manipulate text. NLTK (Natural Language Toolkit) performs fundamental
NLP operations such as tokenization, stop-word removal, and sentiment analysis. spaCy
enhances text processing with advanced features like named entity recognition and part-of-
speech tagging. Modern AI capabilities such as text summarization, keyword extraction, and
topic modeling are implemented using the Transformers library from Hugging Face.
Furthermore, Scikit-learn is employed for machine learning tasks including article
classification, clustering, and personalization of news based on user preferences.

4.2.3 User Interface & API Integration Libraries

For user interface and API integration, the system uses libraries that enable user-friendly
access and interaction. Flask or Django frameworks help build a responsive web interface
through which users can configure preferences and view personalized news. For prototypes
or simpler dashboards, Streamlit provides an interactive and visually appealing way to
present content. Flask-RESTful supports API endpoints that allow modules to communicate
effectively, ensuring the delivery of up-to-date news to users based on their interests.

Conclusion

In conclusion, the system architecture represents a comprehensive and intricate journey from
data acquisition and preprocessing to model selection, training, real-time activity recognition,
and continuous evaluation. Each phase is meticulously designed to enhance the system's
effectiveness, adaptability, and reliability in recognizing a diverse array of human activities.

Dept. of ISE, RNSIT 2025-2026 26

 Chapter 4

Use case diagram

Figure 4.2: Use case diagram

1. Admin

The Admin holds the highest privilege level and is responsible for the overall management
and maintenance of the system. Key capabilities include:

 Monitoring the entire system through Dashboard

 Managing News Categories to organize content

 Reviewing and controlling News Information posted by authors

 Updating and reviewing Author Information

 Managing Subscriber Information such as subscription plans or preferences

 Moderating and managing Comments

 Controlling User Management operations including access and authentication

 Tracking platform usage through Visitor Analytics

 Performing Database Backup to ensure data integrity and disaster recovery

The Admin ensures smooth system operation and security compliance.

Dept. of ISE, RNSIT 2025-2026 27

Chapter 4

2. Author

The Author is responsible for creating and maintaining news content. Their access level is
limited to functionalities relevant to content publication:

Adding, editing, and managing News Information

Updating their Author Profile Information

Viewing and responding to Comments on published news articles

Authors are key contributors in delivering fresh and relevant news to the platform.

3. Subscriber

Subscribers are end-users who consume news content. Their nature of interaction is mainly
passive, but some interactive features are available:

Accessing and reading all available News Information

Posting and interacting through Comments on news articles

Subscribers rely on the system’s recommendation and categorization features for personalized
news delivery.

4.4 Sequence Diagram

Figure 4.3: Sequence Diagram

Dept. of ISE, RNSIT 2025-2026 28

Chapter 4

The sequence diagram of the AI Personal News Reader System represents the flow of
activities carried out between system users and the system components over time. When the
Admin logs into the system, the credentials are verified and upon successful authentication,
access to the Admin dashboard is granted. The Admin can manage news categories, handle
user accounts, monitor visitor analytics, and initiate database backup operations. Each
request issued by the Admin triggers the system to retrieve information from the database,
process updates, and return confirmation, ensuring smooth administrative control.

Similarly, the Author interacts with the system by logging in, where credentials are validated
before granting access to the author panel. Authors can create, edit, and publish news articles.
Once submitted, the system stores the updated news information in the database.
Additionally, Authors can view and respond to comments on their articles, and they can also
update their profile information, which the system saves appropriately.

Subscribers, as end users, interact with the system to read the news. They may log in or
browse directly, after which the system retrieves and displays the requested news content.
Subscribers can also post comments on articles, which the system stores and immediately
updates in the article’s feedback section. Based on user interactions and preferences, the
system may also recommend relevant news articles, enhancing personalization.

Throughout all interactions, the system continuously processes requests, manages

authentication, communicates with the database, and returns results to users according to
their roles and permissions. This logical sequence ensures content publication efficiency,
secure role-based access, and engaging user experience within the AI Personal News Reader
environment.

Dept. of ISE, RNSIT 2025-2026 29

Chapter 4

Chapter 5

IMPLEMENTATION

5.1 Overview of System Implementation

The implementation of the AI Personal News Reader System focuses on a modular design
structure where each functional unit such as News Management, User Management, and
Analytics is developed independently and connected seamlessly. This modularity ensures
easy maintenance, scalability, and faster updates. The system is built with a user-friendly
and intuitive interface, enabling users from different age groups and digital backgrounds
to easily access and interact with content.

1.1.1 ALGORITHMS
1. User Authentication Algorithm
 Compares entered credentials with encrypted values stored in the database.
 Implements session management to keep track of active users securely.
 Prevents brute-force and unauthorized login attempts
2. News Recommendation Algorithm
 Uses Content-Based Filtering to analyze article tags, categories, and user
preferences.
 Optionally uses Collaborative Filtering to suggest news based on trends and similar
user interests.
 Continuously updates recommendations as the user interacts with more content.
3. Search and Sorting Algorithm
 Provides fast keyword-based search results from the news database.
 Sorts news items by latest date, category relevance, or user popularity.
 Ensures minimal delay and proper content organization.

Dept. of ISE, RNSIT 2025-2026 30

Chapter 4

5.1.2. CODE

import os
from flask import Flask, request, jsonify, send_from_directory
from flask_cors import CORS
from dotenv import load_dotenv

from scraper import fetch_news_from_category

from summarizer import summarize_text
from translator import translate_text
from tts import text_to_speech

load_dotenv()

app = Flask(_name_)
CORS(app) # allow calls from file:// frontend

# ---------- HOME ----------

@[Link]("/")
def home():
return "AI News Reader backend is running."

# ---------- GET NEWS BY CATEGORY + SUBCATEGORY ----------

@[Link]("/get-news", methods=["GET"])
def get_news():
try:
category = [Link]("category", "general")
sub = [Link]("sub", "").strip().lower() or None
page = int([Link]("page", "1"))

article = fetch_news_from_category(category, sub, page)

Dept. of ISE, RNSIT 2025-2026 31

Chapter 4

# Build a big text blob for summarization

article_text = f"{[Link]('title','')}. {[Link]('description','') or ''}
{[Link]('content','') or ''}"

return jsonify({
"article": article_text,
"title": [Link]("title", ""),
"source": [Link]("source", {}).get("name", ""),
"published_at": [Link]("publishedAt", ""),
"page": page
})

except Exception as e:
return jsonify({"error": str(e)}), 500

# ---------- SUMMARIZE + TRANSLATE + TTS ----------

@[Link]("/summarize", methods=["POST"])
def summarize():
try:
data = request.get_json(force=True)
text = [Link]("text")
lang = [Link]("lang", "en")

if not text:
return jsonify({"error": "Text is required"}), 400

# 1) Summarize in English
summary_en = summarize_text(text)

# 2) Translate if needed
summary_final = translate_text(summary_en, lang)

Dept. of ISE, RNSIT 2025-2026 32

Chapter 4

# 3) Text-to-speech
audio_rel_path = text_to_speech(summary_final, lang)

# Make full URL for frontend

audio_url = request.host_url.rstrip("/") + "/" + audio_rel_path.replace("\\", "/")

return jsonify({
"summary": summary_final,
"audio": audio_url
})

except Exception as e:
return jsonify({"error": str(e)}), 500

# ---------- SERVE AUDIO FILES ----------

@[Link]("/audio/<path:filename>")
def serve_audio(filename):
audio_dir = [Link]([Link](_file_), "audio")
return send_from_directory(audio_dir, filename)

if _name_ == "_main_":
# backend runs on [Link]:5000
[Link](debug=True)

Dept. of ISE, RNSIT 2025-2026 33

 Chapter 6
TESTING
Testing can be stated as the process of verifying and validating whether a software or
application is bug-free, meets the technical requirements as guided by its design and
development, and meets the user requirements effectively and efficiently by handling all the
exceptional and boundary cases. The process of software testing aims not only at findingfaults
in the existing software but also at finding measures to improve the software in terms of
efficiency, accuracy, and usability. It mainly aims at measuring the specification,
functionality, and performance of a software program or application, so that it may helps in
fixing the bugs.

Test Case is a set of actions executed to verify a particular feature or functionality of

your software application. A Test Case contains test steps, test data, precondition,
postcondition developed for specific test scenario to verify any requirement. The test case
includes specific variables or conditions, using which a testing engineer can compare
expected and actual results to determine whether a software product is functioning as per the
requirements.

6.1 Unit Testing

Unit testing is an essential component of the testing strategy, focusing on validating
individual modules or components of the software. During unit testing, each module is tested
in isolation to verify its internal logic and functionality. This process involves executing test
cases that assess the module's behavior against the specifications outlined during the design
phase. Unit testing helps identify and rectify errors within the boundaries of the modules,
ensuring that each module operates correctly and produces the expected output. By
conducting thorough unit testing, developers can validate the correctness of the code
produced during the coding phase and mitigate the risk of defects and vulnerabilities.

The provided test cases offer specific scenarios and expected outcomes for evaluating
the system's functionality and performance. These test cases simulate various real-world
scenarios, such as the absence of a detected face, side-facing drivers, eye blinking, prolonged
eye closures, and yawning behavior. Each test case is designed to validate specific
functionalities of the system, such as face detection accuracy, eye blink detection, and
drowsiness detection. By executing these test cases rigorously, testers can assess the system's
reliability, accuracy, and responsiveness in detecting driver drowsiness and issuing timely

alerts.

Dept. of ISE, RNSIT 2025-2026 34

Chapter 6

6.2 System Testing

System Testing is a crucial stage in the software development life cycle where the integrated system
is tested as a whole to ensure that it meets the defined requirements and performs correctly in real -
time use. For the AI Personal News Reader System, system testing was performed to validate
functionality, performance, security, usability, and compatibility across different platforms and user
roles. The primary goal of system testing is to detect any defects or issues and ensure that the system
behaves as expected under various conditions.

The testing process was carried out on all essential modules such as User Login and Authentication,
News Posting and Management, Personalized Recommendation Display, Comment Functionality,
Visitor Analytics, and Database Backup. Each feature was tested from the perspective of Admin,
Author, and Subscriber roles to ensure proper access rights and functionality. Different types of tests
including functional testing, performance testing, user acceptance testing, and security testing were
conducted to evaluate the overall quality of the system. During the testing phase, identified issues
were documented and resolved to improve system reliability and user experience.

As a result, the system was verified to be stable, accurate, and responsive, successfully fulfilling the
expected requirements. The AI-based news recommendation functions correctly by presenting
relevant content to users, and all system actions such as login validation, content retrieval, comment
submission, and analytics display were confirmed to operate smoothly. Overall, system testing
ensured that the AI Personal News Reader System is ready for deployment, fully capable of
providing a secure, personalized, and user-friendly digital news platform.

System testing was conducted to evaluate the AI Personal News Reader as a complete and integrated
application. The purpose of this testing was to ensure that all functional requirements, such as
delivering personalized news based on user preferences, supporting multiple news categories, and
integrating the AI personalization engine, were working correctly. Additionally, non-functional
aspects like performance, reliability, and responsiveness were assessed to ensure the system could
handle multiple requests efficiently and provide a seamless user experience. During system testing,
the application was tested end-to-end, simulating real-world usage scenarios, which helped identify
and resolve any discrepancies between expected and actual behavior. This comprehensive testing
ensured that the overall system met the specified requirements and was ready for deployment.

Dept. of ISE, RNSIT 2025-2026 35

Chapter 6

Table 6.1: Test Case Daigram

6.3 Validation Testing
Validation is an essential process in software development that involves checking whether
the software product is up to the mark and meets the high-level requirements. It is a crucial
step in ensuring that the developed software is the right product and aligns with the user's
needs. The process of validation involves checking the actual product against the expected
product, ensuring that it meets the intended functionality and specifications. The primary
goal of validation is to determine whether the developed software satisfies the intended use
and the customer's requirements.

Validation is a dynamic testing process that checks the behavior of the software in a
dynamic environment. Usability and accessibility aspects were also considered, verifying
that the interface is intuitive, navigation is straightforward, and users can easily interact
with the application. Performance validation ensured that news fetching and AI
personalization were fast and responsive, even under higher loads. Additionally, feedback
from test users was collected and analyzed to ensure that the system's output aligns with
their expectations. Through this comprehensive validation process, the AI Personal News
Reader was confirmed to be fit for purpose, reliable, user-friendly, and capable of
providing an effective personalized news experience, meeting both functional and user
satisfaction criteria.

Dept. of ISE, RNSIT 2025-2026 36

Chapter 7
DISCUSSION OF RESULTS
7.1 Summary
The AI Personal News Reader was developed and tested to provide a personalized, user-
friendly, and efficient news reading experience. Through comprehensive testing—including
unit, integration, system, performance, usability, security, regression, and validation testing—
the system’s functionality and reliability were thoroughly evaluated. The results demonstrated
that the application successfully fetches news from multiple sources, filters and categorizes
content according to user preferences, and delivers personalized news summaries in a timely
and accurate manner.

System testing confirmed that all modules worked together seamlessly, while performance
testing showed that the application could handle multiple requests without significant delays.
Usability and validation testing indicated that the interface is intuitive and meets the
expectations of end users. Security testing ensured that user data, such as preferences and
activity, is handled safely. Overall, the results indicate that the AI Personal News Reader meets
its intended objectives, providing a robust, efficient, and satisfactory experience for users
seeking personalized news.

7.2 Results and Snapshots

Fig 7.1: login page

Dept. of ISE, RNSIT 2025-2026 37

Chapter 7

Fig 7.2: Home page

Fig 7.3: News summarization

Dept. of ISE, RNSIT 2025-2026 38

Chapter 8

CONCLUSION AND FUTURE WORK

8.1 Conclusion
The AI Personal News Reader system successfully addresses the growing issue of
information overload by providing a smart, personalized news browsing experience. By
integrating advanced technologies such as Natural Language Processing and Machine
Learning, the system is capable of filtering, categorizing, summarizing, and
recommending news content tailored to individual user preferences.

Through its automated data aggregation and real-time updates, users are ensured access
to the most relevant and latest news without the need to manually search multiple
platforms. The system’s user-friendly interface enhances accessibility and improves the
overall reading experience by presenting concise and meaningful information
efficiently.

This project demonstrates how artificial intelligence can significantly transform modern
digital news consumption, saving time and improving engagement. In the future, this
system can be enhanced further by including multilingual news support, sentiment-
based delivery, voice-assisted reading, and stronger user-behavior modeling to improve
recommendation accuracy. Overall, the AI Personal News Reader stands as a valuable
innovation in simplifying and personalizing the way people consume news in today’s
fast-paced digital world.

8.1 Future Enhancements

The AI Personal News Reader system has significant potential for future improvement
and expansion. One key enhancement includes extending multilingual support, enabling
the system to collect, translate, and summarize news from various regional and
international languages to cater to a broader audience. The integration of voice-assisted
features using text-to-speech technology can further enhance accessibility, allowing
users to listen to news updates while multitasking. Additionally, advanced sentiment
analysis can be incorporated to enable filtering of news based on emotional tone, helping
users prefer content aligned with their mood or interest.

Dept. of ISE, RNSIT 2025-2026 39

 REFERENCES

[1] A Real-time Driving Drowsiness Detection Algorithm With Individual Differences

Consideration Feng You 1, Xiaolong Li 1, Yunbo Gong 1, Haiwei Wang 2, And Hongyi
Li
[2] Facial Expression Recognition Based on Attention Mechanism Jiang Daihong , Hu
yuanzheng, Dai Lei, and Peng Jin Xuzhou University of Technology, College of
Information Engineering, Xuzhou 221000, China
[3] Stefan Oniga, József SütĘ “Human activity recognition using neural networks” 2014
15th International Carpathian Control Conference (ICCC) Island Loss for Learning
Discriminative Features in Facial Expression Recognition: Jie Cai* Zibo Meng* Ahmed
Shehab Khan Zhiyuan Li Yan Tong Department of Computer Science and Engineering,
South Carolina University
[4] A 3-Dimensional SIFT Descriptor and its Application to Action Recognition : Paul
Scovanner, Computer Vision Lab University of Central Florida, Saad Ali
[5] Identity-Aware Convolutional Neural Network for Facial Expression Recognition:Zibo
Meng, Innopeak Technology [Link] Liu,University of South Carolina,Jie Cai,
University of South Carolina,Shizhong Steve Han,Qualcomm,Yan Tong
[6] Design of Real-time Drowsiness Detection System using Dlib, Authors: Conference:
2019 IEEE International WIE Conference on Electrical and Computer Engineering
(WIECON-ECE) Shruti Mohanty,Shruti Hegde,Supriya PrasadJ, Manikandan
[7] Recognizing Facial Expressions in Image Sequences Using Local Parameterized
Models of Image Motion, MICHAEL J. BLACK, Xerox Palo Alto Research Center,
3333 Coyote Hill Road, Palo Alto, , 1995;
[8] Automatically Detecting Action Units from Faces of Pain: Comparing Shape and
Appearance Features Patrick Lucey, Jeffrey Cohn, Simon Lucey,Robotics Institute,
Carnegie Mellon University Pittsburgh, PA, 15213 USA.
[9] Facial Expression Analysis Ying-Li Tian,1 Takeo Kanade2, and Jeffrey F. Cohn2,3 1
IBM T. J. Watson Research Center, Hawthorne, NY 10532, USA. yltian@[Link]
2 Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213, USA.
tk@[Link] 3 Department of Psychology, University of Pittsburgh, Pittsburgh, PA
15260, USA.
[10] Towards an Example-Based Image Compression Architecture for Video-Conferencing
,Sebastian Toelg and Tomas Poggio

Dept. of ISE, RNSIT 2025-2026 40