Matchmaking Beams with Ads

We put heads together on Tue, Nov 11, 2007 and hashed out a strategy for roll out of BeamAds. While it was decided to build support for RON (Run of Network) campaigns first of all, the process of matchmaking Beams with Ads was hashed out well. These decisions and explanations are iterated in some detail below.

The pros and cons of keyword matching versus categorization took center stage.

Mac pointed out that looking to the past would reveal a consistent trumping of categorization by keyword matching, i.e. keyword matching on www.google.com, www.yahoo.com and www.live.com making categorized directories such as www.dmoz.org, directory.google.com, dir.yahoo.com and www.excite.com irrelevent.

Brad felt Advertisers may prefer the ease of selecting just a few top-level Categories, as opposed to specifying a few hundred keywords in hopes of making frequent matches.

After investigating www.adbrite.com and the 200lb gorilla adwords.google.com along with a few smaller fish, consensus was reached:

Summary
Allow advertisers to manually enter and edit keywords as well as select from single-level categories of keywords to add in bulk.

In brief, Publisher Keywords are farmed and extrapolated. Advertiser keywords are read and extrapolated. A lite matchmaker runs between the long list of extrapolated keywords on both sides. All done.

No, this is not a simple read of Beam Tags. The goal here is to generate a reasonably long list of keywords from all the data we hold at the time a Beam is requested. Keywords are farmed from the following sources:

• Beam Text
• Beam Tags
• Publisher’s URL
• GeoIP location of end user

Keywords are simply read from an Advertiser’s list. Remember the keyword selection tool itself is to provide one-click operations to add entire categories of keywords, if desired.

Applied to both Publisher and Advertiser keywords, extrapolation is how we make matches when none are found, and will eventually contribute to ranking multiple matches by contextual relevance.

The criteria for extrapolation are straight-forward:

• Synonyms
• Word association

The latter will be a constantly evolving, automated function of BeamMe with potential input from online repositories, smart scripting, word games, etc. The former will be the result of interaction with a good thesaurus API.

• Publisher has defined Beam Tags: custom paint jobs, cars, pimp my ride, chrome, spray paint
• Ad Campaign is targeted with keywords: household insurance, vehicle insurance, travel insurance
• There are not many Ad Campaigns running through BeamMe yet

Why does this work?
Beam Tags cars and ride are both extrapolated to vehicle. If this partial match to Ad Campaign keyword vehicle insurance is the best we have, it takes precedence over Run of Network campaigns.

• Publisher has no Beam Tags defined. There are no significant words in Beam Text.
• Ad Campaign is targeted with keywords: finance, credit cards, aussie, b2b, lending
• User is in Sydney

Why does this work?
GeoIP farms Australia and NSW (along with whatever else it can find) as keywords. There are extrapolated to many more keywords through synonyms and word association, including aussie, which returns a positive match.