sayyedabrarakhtar.com.np
theme
~/portfolio

// ACADEMIC CONTENT PORTAL · Computer Science

Thesis Topic Ideas: Nepali Fake News Detection Using NLP

Level: Bachelor, MasterDifficulty: Advanced★ Popular Choice

🎓 Academic Journey Progression Map

1. Introduction & Problem Statement

Overview: Develop an NLP classifier trained on Devanagari news text datasets to detect sensationalized and fake news articles across Nepali digital media portals.

Background Context (Nepal): The proliferation of unverified news portals and social media clickbait in Nepal highlights the need for automated Devanagari text classification tools.

2. Research Objectives

  • Curate and label a dataset of 3,000+ authentic and deceptive Nepali news headlines and articles
  • Implement Devanagari text preprocessing (stop-word removal, stemming, tokenization)
  • Train TF-IDF + XGBoost and Devanagari BERT (NepBERT) NLP models
  • Achieve 85%+ F1-score on news veracity classification
  • Deploy web browser extension or demo portal flagging suspicious articles

3. Proposed Methodology

  1. Scrape news articles from verified and unverified online news portals in Nepal
  2. Preprocess Devanagari text using NLTK / Stanza NLP pipeline tuned for Devanagari script
  3. Extract linguistic features (hedging words, exclamation density, sentiment polarity)
  4. Train and compare Logistic Regression, Naive Bayes, NepBERT transformer models
  5. Evaluate precision, recall, confusion matrix, and inference speed

$ Worked Example / Sample Scenario

Sample Scenario: A user pastes a viral news headline into the web portal. The NepBERT model analyzes Devanagari syntactic structure and flags 'Fake News Risk: 92%' due to sensationalized phrasing and missing source attribution.

4. Thesis Chapter-by-Chapter Outline

Chapter 1: IntroductionTU/KU/NEB standard

Background on digital news expansion in Nepal, clickbait problem, and research goals

Chapter 2: Literature ReviewTU/KU/NEB standard

Devanagari text processing, transformer models (NepBERT), and fake news detection literature

Chapter 3: Research MethodologyTU/KU/NEB standard

Dataset annotation rules, feature engineering, and model architecture design

Chapter 4: Experiments & ResultsTU/KU/NEB standard

Model performance comparison, ROC curves, F1-scores, and error analysis

Chapter 5: Conclusion & DeploymentTU/KU/NEB standard

Summary, browser extension demo overview, and future dataset expansion

5. Recommended Tools & Technologies

To implement the practical, technical, or analytical portions of this thesis topic, the following software tools, libraries, or APIs are recommended:

PythonPyTorch / HuggingFaceNepBERTScikit-LearnFastAPIReact

6. Core References & Academic Sources

  • [1]NepJOL — Devanagari NLP and Text Classification Research Papers <!-- VERIFY: needs confirmation -->
  • [2]IEEE Transactions on Computational Social Systems <!-- VERIFY: needs confirmation -->

7. Frequently Asked Questions (FAQs)

Q: Where can I get a Devanagari news dataset in Nepal?

You can scrape public articles from major news portals (e.g., Ekantipur, OnlineKhabar) and annotate a benchmark set with guidance from your advisor.

Related Academic Topics & Student Tools

available for workKathmandu, Nepal 🇳🇵contact@sayyedabrarakhtar.com.np