RevGuard

Inspiration

We noticed that many online location reviews are misleading, irrelevant, or spammy, which makes it hard for users to trust them. We wanted to create a tool that helps users make informed decisions while exploring AI further.

What it does

RevGuard leverages machine learning and NLP to ensure that online location reviews are trustworthy, relevant, and policy-compliant. By automatically detecting spam, promotional content, and irrelevant reviews, RevGuard helps users make better decisions, protects businesses from unfair misrepresentation, and empowers platforms with automated, reliable moderation.

How we built it

We collected raw data of Google location reviews and weakly labelled them according to quality and relevancy. Then, we sampled out a gold set in which we manually labelled it to ensure accuracy for evaluation and testing. We engineered features using text embeddings and TF-IDF, then trained classifiers like Random Forest and MLP. We also applied heuristics to enforce policy rules automatically.

Challenges we ran into

Differentiating genuine negative reviews from spam or irrelevant content.
Balancing precision and recall to avoid over-flagging legitimate reviews.
Understanding the pipeline and executing it

Accomplishments that we're proud of

Built a working ML pipeline capable of filtering out low-quality and irrelevant reviews.
Achieved good enough performance on validation data with high precision for spam detection.
Demonstrated a scalable approach that could be deployed on real review platforms.

What we learned

NLP feature engineering (TF-IDF, text embeddings) is critical for understanding nuanced review content.
Effective labelling and heuristics significantly improve model performance.
Balancing automated detection with fairness to users and businesses is key.

What's next for RevGuard

Enhancing the performance of our model
Incorporate more advanced language models for better semantic understanding.
Extend to multi-lingual review support.
Integrate a real-time dashboard for platforms to monitor review quality and trends.