The problem
A five-star product with 3 reviews outranks a 4.7-star product with 2,000 reviews on most e-commerce platforms. Raw average ratings are trivially manipulated, ignore recency, and throw away the richest signal available — what reviewers actually wrote.
My undergraduate thesis (supervised by Dr. Farida Chowdhury, Professor, CSE, BRAC University) tackled this as an HCI + applied data science problem.
The approach
Three components combine into one adaptive score:
1. Bayesian averaging. Every product's score is pulled toward a prior (the category mean) with strength inversely proportional to its review count. A 5.0 from 3 reviews gets shrunk hard; a 4.7 from 2,000 reviews barely moves. This makes scores statistically stable and manipulation-resistant.
2. Exponential time decay. Reviews are weighted by recency, so a product that declined in quality two years ago can't coast on old praise. The decay rate was tuned by hypothesis testing against held-out rating trajectories.
3. AI sentiment extraction. Review text is run through sentiment analysis, and the extracted signal dynamically adjusts the predefined rating weights — a glowing 4-star review and a bitter 4-star review stop counting the same.
My contribution
I was solely responsible for all data-related work: exploratory data analysis, hypothesis testing, the algorithm design, and statistical validation. The thesis supervisor explicitly commended the originality and statistical rigour of the rating system contribution.
From thesis to product
The algorithm didn't stay in the paper. I built a Chrome extension that applies the adaptive rating directly on live e-commerce product pages — recomputing the displayed score from the visible review data, so shoppers see the robust rating next to the platform's naive one.