Ever go to a store such as Amazon and search for something and try to find the best rated product? I find that the rating sorting doesn’t accurately display products by consumer approval. Just having a high product rating shouldn’t list a product first when sorting by rating, especially if the product has 5 out of 5 stars rated by only one user.  And it seems that if only 1 of 3 people even thought it was a decent review that those 5 stars are not as reliable as they you would want them to be.  Sure you can blame the people who write them and then rate every review yourself, and then sort the reviews for each product by the most helpful reviews, but that’s a lot of work.  I feel that a more automated system that would add weights to each star rating and incorporate that weight to the total number of reviews and number of helpful reviews as well could acheive a formula that would allow for a more relevant list of products ordered by a logical rating that takes into account the quality of the review.

I played with a few formulas before I arrived at this one, which I will state isn’t perfect, but you can get the idea.  And of course what you use as the weight of the ratings and how you apply the helpfulness will affect how many points are finally added to each product.

POINTS = v * s + (y-n) * s
v: number of reviews
s: curved scale value for rating based on  5=25, 4=10, 3=1, 2=0.01, 1=0.005
y: number of helpful review votes across all reviews
n: number of un-helpful review votes across all reviews

I did a search for “802.11n router” on Amazon and found the top six products ordered by user rating had a 5 star product with 1 review listed first.  After my system this is how those top six should be listed

Amazon Rank	My Rank	Rating	Reviews	Helpful Votes	Total Votes	My Points
5.	1.	4.0	344	1232	1663	11450
6.	2.	4.0	119	567	793	4600
2.	3.	4.5	12	34	41	780
3.	4.	4.5	3	11		240
4.	5.	4.0	4	0	0	40
1.	6.	5.0	1	1	3	0

Of course items that are older will have more reviews, but at the same time these items will also have reviews that will be rated by others that will properly let you know if something is truly rated high, not just the highest rated. Other possibilites are to make the formula more specific by multiplying the helpful scores by the rate they are marked for and not simply adding all the thumbs ups and thumbs down collectively as I have.

Other considerations could even be taken involving the person that gave the rating. These could be extra points given to a rating by a person that has had a certain level of helpful reviews on other products, by this that could mean that a person that gives a 3 rating to a review would then possibly count as multiple 3s based on his level. In this example if “Bobby M.” reviewed 20 products all of which have received more than 100 helpful reviews with a 90% helpful rating then his vote could be counted as 5 votes instead of one as he is considered an “expert.”

I think this is a good start into creating a more helpful way to present product ratings especially in a world were sorting and finding the perfect product is very difficult as more and more products flood the market.