Recommendation suggestion
Saurabh Nagrecha
I've seen several feedback items here about recommendations being not exactly on the mark, and I agree with that.
EXAMPLE:
I'll add another example here where I was on Amazon and I was looking for
"Starbucks Italian Roast Dark Roast Whole Bean Coffee, 12 Ounce (Pack of 6)"
and the recommendation was for
"Fivesso Relax Me Coffee Soaps: 3-Pack"
POSSIBLE CAUSE:
What this shows me is that there is a certain "product name on page" to "best matched inventory items on Ethically" search going on. Sure enough, both contain the word "coffee" but they're not the same thing.
SOLUTION(s):
* If the best match scorer doesn't find a good quality match, don't return anything. Business tradeoffs apply - is the cost of showing a bad match worse than showing "something"?
* Change the best-match scorer. Current matcher seems to be based on string token similarity, which is a great starting point, but there are several alternatives around. Depends on dev bandwidth to swap these components out.
Some excellent implementations I love are:
etc
Jada McLean
Thank you so much for the feedback, Saurabh! This is very helpful and we will explore these options
Saurabh Nagrecha
Another source of improvement could be to change the matching algo from:
* product_on_webpage <-> [product_title_1, product_title_2, ...], to
* product_on_webpage <-> [{product_title_1, product_description_1}, {product_title_2, product_description_2}, ...]
The algos are largely similar ... they'll just get a bit more nuanced. You only need to update the database of Ethically-onboarded products whenever the partner company joins Ethically with whatever products they have to offer.