About & Methodology
Understanding the technology and metrics behind the ShelfDB Discovery Engine.
Why ShelfDB was built
Finding a truly great book online has become increasingly difficult. Popular bestseller lists are often influenced by publisher marketing budgets, and raw average ratings on review platforms can be easily skewed. A book with a perfect 5.0 rating from just three reviews will frequently rank above a classic book with a 4.2 rating from fifty thousand reviews.
ShelfDB was built to solve this. By combing a granular filter engine (allowing you to search by length, language, keywords, and release year) with a mathematical sorting algorithm, we seek to surface high-quality, genuine books that are worth your time.
The Weighted Rating Algorithm
To create a fair rating that balances both the average score and the volume of reviews, ShelfDB employs a Bayesian Average (similar to the rating formula used by IMDb for their Top 250 list).
Bayesian Weighted Rating (WR) Formula:
By introducing the global dataset average ($C = 3.89$) and a minimum threshold ($m = 50$ ratings), the formula pulls books with very few ratings towards the average. Only books that maintain a high rating across a significant number of reviews can achieve a high Weighted Rating. This successfully filters out low-volume outliers while surfacing highly popular classics and hidden gems.
The Data Source & Gaps
Our catalog is built from the Hardcover book database and currently spans 100,000+ titles, with more added regularly as we grow toward launch. Metadata — covers, ratings, series, and descriptions — reflects each book's record in Hardcover at the time it was indexed, and is refreshed periodically rather than live.
In accordance with our data hygiene guidelines:
- Books with missing page counts or languages were excluded from the database to ensure search results are 100% accurate.
- HTML markup inside descriptions was cleaned to improve readability.
- A rule-based keyword classifier was used to map books to standard genres based on their titles and descriptions.
Full Affiliate Disclosure
ShelfDB is built with a focus on original content and high-quality utility. We monetize the platform using a hybrid affiliate model with book retailers including Amazon Associates, Bookshop.org, and ThriftBooks. When you click one of our buttons to purchase a book or check its price, a referral cookie is set in your browser, and we receive a small commission of the sale (typically 4–10%) from the retailer at no cost to you. As an Amazon Associate I earn from qualifying purchases.
In order to remain compliant with affiliate agreements:
- No Hardcoded Prices: No currency values are stored in our database or displayed on this site. Price checks are routed dynamically.
- Used & Independent Alternatives: We offer alternative links for Bookshop.org (supporting independent local bookstores) and ThriftBooks (offering used/second-hand editions) alongside Amazon.
- Trust & compliance: The site operates independently of the retailers and does not scrape or display live inventory.