Natural Language Processing script that can be "taught" how to match products

We have two datasets, which are made of text strings. They are similar but not exactly the same. Humans can distinguish which are correct matches. Project is to build a system that can learn from the matches we approve vs. decline. We can effectively limit possible matches to about 10 products. Only a few are correct matches. We need a system that can be taught through example which are correct matches and which are not. For example: Aveeno Products, Shampoo or Conditioner, Hair Styling or Skincare Potential Matches: 1.) Aveeno Facial Care Product, excludes trial sizes and cleansing bars 2.) Aveeno Facial Care Product, excludes trial sizes and cleansing bars 3.) Aveeno Positively Ageless Skin Strengthening Body Lotion or Hand Cream (IE) 4.) Aveeno Active Naturals, sunscreen product ets 5.) Aveeno Sun Care Item, TARGET coupon 6.) Aveeno Hand, Body, or Baby Lotion, excludes trial and 2.5 oz 7.) Aveeno Body Wash or Haircare Prodcut, excludes trial 8.) Aveeno Facial Care or Suncare Product, excludes trial and cleansing bars 9.) Aveeno Hair Care, Videos Available to Watch until 9/1/2012 In this example, 1, 2, 3, 6, 7, 8, are matche