Sunday, August 9, 2009

Polysemy, ambiguity and spam recipes

Once in a while, I check my mail sent to my Gmail account using the online interface. I generally go directly to the spam folder to make sure that the spam filter has not been over-zealous, and sometimes just for laughs.

Some advertising makes it through my default Adblock Plus config but what keeps me from blocking the frame is the (certainly limited but not non-existent) entertainment value of this sponsored links section. Most of the text ads deemed "contextually" relevant to the mail classified as spam have to do with unsolicited mail. Some, however, are pretty much off target and that makes them way more interesting: the relevance computation is evidently based on "spam" construed as gelatinous mystery/porky meat.
Spam Confetti Pasta
French Fry Spam Casserole
Spam Vegetable Strudel
Spam Veggie Pita Pockets
Spicy Spam Kabobs
Savory Spam Crescents
Spam Primavera
Spam Imperial Tortilla Sandwiches
Spam Fajitas
Spam Skillet Casserole
Spam Quiche
Spam Hashbrown Bake
Ginger Spam Salad
Yikes.

Resolving ambiguity originating in polysemy is no easy task. And yet, one could assume that the engine used by AdSense to process keywords and calculate ad relevance would know what "spam" means when it is a label associated by Gmail itself to an electronic message, don't you think?

No comments:

Post a Comment