It seems that a lot of user generated content on the internet has mutually exclusive sorting parameters. This makes finding the good stuff pretty hard. For example, go to a site like rateitall.com, allrecipes.com, or newegg.com and do a search. These search results can then be sorted by several parameters, but usually only one sort can be performed at a time. Most often, two of the parameters are user rating and popularity. With rating, users rate the item/product/whatever it is in an “x out of 10″ manner, while popularity lists the number of times the item or product has been rated/viewed/purchased.
This leads to a problem: sorting by rating may yield results that have been highly rated by a small number of users. With a small sample size, there is a really high potential for bias. Sorting chicken recipes by user rating may yield a top result with one user rating of 10/10, but the recipe makes use of a ton of garlic. It just so happens that this one user really likes garlic, hence the high rating. However, if many people were to rate this recipe, due to the excessive garlic it might fall to an average rating of 3/10. The results of the rating-based sort would be completely altered with much larger sample sizes. Conversely, if you have weird tastes, you might prefer something that most people detest and vice versa, making rating-based sorts with large sample sizes useless.
One other rating problem is that of user satisfaction. If a user really has trouble with a particular product or really enjoys something, he or she is more likely to express that opinion than if it was something mediocre. This will really affect the distribution of ratings for items/products as most ratings will be either really good or really bad, but very few will be in the 50 percentile range. Forcing each visitor/user/customer to rate each item may solve this problem, but could be a real pain for those who hate leaving feedback at all. I’ll leave this one alone for now.
On the flip side, sorting by popularity alone may also yield poor results. The most popular items or products may have mediocre ratings. Take chain pizza, for example. Everyone knows Domino’s, Pizza Hut, and Papa John’s, but few would argue they make the best pizzas. The question becomes one of how to use rating and popularity to generate meaningful results — specifically results that cater to your tastes.
The quickest approach is to combine rating and popularity into one sorting mechanism. Results can be sorted by rating but weighted by popularity. Thus, a singly-rated item of 10/10 would fall behind a very popular item with a rating of 9/10. Of course, the question remains of how to determine the weight. The issue of individual taste is also another matter. The best way to solve this would be to further constrain the search parameters based on those tastes, weeding out the stuff you don’t like while placing greater emphasis on what you do like.
In short, it’s hard to know if your best search results are really what you are looking for. Often, the best choice for you is buried deep somewhere and will never be found.
Recent Comments