Fake News Detector

Inspiration

My inspiration for this project came while scrolling through Instagram Reels. I noticed a overwhelming amount of misinformation around elections, political figures, and major events. Fake news thrives on triggering emotional responses, leaving people feeling helpless, indifferent, or detached. When we are constantly exposed to high-tension content, it strips away emotional weight from genuine, real-world problems. I wanted to build a tool that helps restore human empathy and trust in society and governance.

What it does

This model will aim to differentiate between fake news and real news via inserting a link into the program and returning whether it's fake news or not and the percent probability of it being fake news or not.

How I built it

Data Preprocessing: I retrieved a dataset from Kaggle containing news articles labeled as fake or real, complete with metadata like title, author, full text, and publication date. Using Jupyter Notebook, I cleaned the dataset by dropping null values, standardizing the text, and stripping stop words (e.g., "the", "an", "at").

NLP Pipeline: I processed the raw text using spaCy for tokenization and lemmatization.

Feature Extraction: I generated numeric word vector representations using TF-IDF and implemented document embeddings to represent full passages.

Model Training & Deployment: Using Scikit-Learn, I split the data, trained a Logistic Regression classifier, and evaluated its accuracy. I then integrated BeautifulSoup4 to handle web scraping so the pipeline can fetch live text directly from a provided URL and feed it into the model for predictions.

Challenges I ran into

I initially struggled with proper data cleaning techniques and live web scraping. Because these concepts were new to me, I relied heavily on library documentation and targeted YouTube tutorials to master BeautifulSoup4 and refine my preprocessing pipeline.

Accomplishments that i'm proud of

I'm proud of finishing this project because I was coming at this with little or no real world knowledge on how to parse and clean the data and to train the model since all my knowledge was theoretical so far but during this datathon, I was able to find out and being able to research deep into AI training while still being able to apply it.

What I learned

I learned out to clean the data, steps to take, how to train, test, and split the data, and how to operate web scrapers like beautifulsoup4.

What's next for Fake News Detector

The next step for the Fake News Detector is to begin analyzing the transcript of Youtube videos and tiktoks to deem whether a said video or tiktok contains fake news or not.

Built With

• dataset
• datathon
• fakenews
• python
• solo