Bayesian rating system with multiple categories for each rating

I'm implementing a rating system to be used on my website, and I think the Bayesian average is the best way to go about it. Every item will be rated in six different categories by the users. I don't want items with only one high rating to shoot to the top though, which is why I want to implement a Bayesian system.

Here is the formula:

Bayesian Rating = ( (avg_num_votes * avg_rating) + (this_num_votes * this_rating) ) / (avg_num_votes + this_num_votes)


Because the items will be rated in 6 different categories, should I use the average of the sums of those categories as "this_rating" for the Bayesian system? For instance, take one item with two ratings (scale of 0-5):

Rating 1:
Category A: 3
Category B: 1
Category C: 2
Category D: 4
Category E: 5
Category F: 3
Sum: 18

Rating 2:
Category A: 2
Category B: 3
Category C: 3
Category D: 5
Category E: 0
Category F: 1
Sum: 14


Should "this_rating" be simply the average of the sums listed above? Is my thinking correct, or should a Bayesian system be implemented for each category as well (or is that overthinking it)?

1 Answer

It depends on whether you want to wind up only with a cumulative rating of each object, or category-specific rating. Having a separate system in each category sounds more realistic, but your particular context might suggest otherwise. You could even do both a category-specific and overall rating!

• I agree. Also, depending on the domain a weighted composite of categories might be a more appropriate index of an overall rating. – Jeromy Anglim Aug 19 '10 at 10:55
• Thanks for the thoughts Aniko. I'll take a look at this in the next few days. – James Skidmore Aug 20 '10 at 6:18