🧠 TrustRank: A Smarter Way to Score Product Reviews

By Aarathisree Balla

❓ What’s Wrong with Average Ratings?

Every time you shop online, the little ⭐ rating next to a product helps you decide whether to buy it. But have you ever wondered:

• Are those 5-star reviews real?
• Is a 4.9 rating from 3 years ago still reliable?
• Do 1-star reviews with no helpful votes really matter?

Traditional average rating systems are too simplistic. They treat every review equally, ignoring important factors like helpfulness, recency, and tone.

💡 Introducing TrustRank

TrustRank is a trust-aware scoring algorithm I built to improve how we rank and interpret product reviews. Instead of just taking the mean star rating, TrustRank calculates a smarter score using:

• NLP-based Sentiment: Was the review positive or just the rating?
• Helpfulness Votes: Did other users find this review useful?
• Time Decay: How recent is the review?
• Star Rating: Still important, but not the only factor.

TrustRank = 0.4 × Rating + 0.2 × Sentiment + 0.3 × Helpfulness + 0.1 × Time Decay

🧪 The Results

To test TrustRank, I created a sample dataset of diverse product reviews. When compared to average star ratings:

• Spearman Rank Correlation: 0.93 — Significantly better alignment with trustworthy product rankings
• Top-3 Overlap: Products ranked by TrustRank showed stronger consensus with user feedback
• Disagreement Detection: TrustRank flagged items with suspiciously high ratings but poor sentiment or outdated reviews

📈 Visualizing the Difference

I built a Streamlit app that:

• Computes TrustRank for any dataset
• Shows visual comparison with average ratings (bar chart, scatter plot, box plot)
• Highlights mismatches and bias in traditional scores

This isn’t just about ratings — it’s about building trust into ranking systems.

Check it out here 👉: https://trustrank-algorithm.streamlit.app/

🔧 Tech Stack

• Python, Pandas, NumPy
• VADER Sentiment Analyzer
• Matplotlib, Seaborn
• Streamlit for interactive frontend

Github Repository : https://github.com/Aarathi1535/TrustRank-Algorithm

🌍 Why It Matters

Platforms like Amazon, Flipkart, and Yelp rely heavily on user-generated reviews. But without trust-aware algorithms, these systems can mislead customers and damage credibility.

TrustRank reflects my effort to design ML solutions that address practical challenges in review credibility and user trust, using sentiment, helpfulness, and recency as core signals.

📄 Research Status

This project represents an early-stage initiative as I explore meaningful ways to contribute to research in review ranking and trustworthy AI. While not yet published, TrustRank is a foundational step in my effort to build interpretable, real-world machine learning solutions.

💬 Share Your Thoughts

Have thoughts, suggestions, or critiques on the approach? I’d love to hear your perspective — whether you’re into NLP, recommender systems, product trust metrics or just a student with curious mind. Drop your opinion in the comments or reach out directly!

🙋‍♀️ About Me

I’m Aarathisree Balla, an engineering student and research enthusiast passionate about NLP, trustworthy AI, and solving impactful problems using applied machine learning.

Let’s connect: LinkedIn