Most local marketplace feeds suffer from a structural flaw: chronological bias. By sorting listings purely by the time they were created, platforms force vendors into a cycle of constant reposting. Genuinely high-quality storefronts with history and reviews get pushed down by low-effort spam that posted minutes ago.
The Fallacy of Simple Averages
Simply sorting listings by average rating (e.g., stars) is equally flawed. A listing with one 5-star review will outrank a listing with fifty reviews and a 4.9-star rating. This is mathematically dishonest and discourages vendors from building long-term reputations.
Weighted Bayesian Estimations
To solve this, Groveh implements a Bayesian rating model. The formula adjusts the raw average rating of a listing based on the volume of reviews it receives relative to the platform average:
W = (R * v + C * m) / (v + m)
Where W is the weighted rating, R is the listing's average rating, v is the number of reviews, m is the threshold minimum reviews required to rank (we set this dynamically), and C is the average rating across the entire marketplace. This ensures that listings with deep, positive history naturally float to the top, while new listings are gracefully benchmarked against the average.
Geospatial Re-ranking
Once the quality scores are computed and cached, we layer on geospatial proximity using Uber's H3 hexagonal indexing. Proximity acts as a regional filter, rather than a raw sort key. This ensures that you are shown the best-reviewed vendors in your local hexagon tier, instead of just the absolute closest seller regardless of their quality.