Weibo Fake News Detection (UE23CS352A Mini Project)

Team Members

• Kavana H – SRN: PES1UG23CS292
• Keshav Singhal – SRN: PES1UG23CS299

---

Problem Statement

The goal of this project is to detect real vs fake news in Chinese Weibo microblog posts using machine learning and deep learning models.

---

Approach

We implemented and compared several models:

• Naive Bayes
• Logistic Regression
• Support Vector Machine (SVM)
• Artificial Neural Network (ANN) (using TensorFlow)

Preprocessing Steps

• Tokenization using jieba
• Stopword removal
• TF-IDF vectorization (for classical ML models)
• Sequence embedding (for ANN model)

---

Dataset

• Weibo Microblog Posts Dataset
• Dataset Link: Weibo Dataset on GitHub

---

Results

Model	Accuracy
Naive Bayes	0.862
Logistic Regression	0.853
SVM	0.857
ANN	0.8213

---

Setup Instructions

1️⃣ Clone the Repository

git clone https://github.com/Kavana-coder/Weibo-Fake-News-Detection
cd Weibo-Fake-News-Detection


2️⃣ Set Up Python Environment

Ensure you have Python 3.8+ installed.

pip install -r requirements.txt

Running the Project
Step 1: Data Preprocessing & Model Training

Run the main notebook:

jupyter notebook main.ipynb

This notebook will:

Load and clean the Weibo dataset

Perform text preprocessing (tokenization, stopword removal)

Train Naive Bayes, Logistic Regression, SVM, and ANN models

Automatically save trained model files:

svm_model.pkl

tfidf_vectorizer.pkl

ann_model.h5

tokenizer.json

results.csv

Step 2: Generate Results

After training, the notebook generates results.csv containing predictions from all models.

Step 3: Run the GUI Application

Launch the Tkinter GUI for real-time predictions:

python gui_app.py

GUI Features:

Enter a Weibo post text

Predict Real or Fake using both SVM and ANN

Display predictions instantly

Notes

The following files are required for GUI predictions:

svm_model.pkl,

tfidf_vectorizer.pkl,

ann_model.h5,

tokenizer.json,

results.csv.

These files are generated automatically after running main.ipynb.