It’s not usually good to take too strong of a position on any topic, but the more I see what’s termed contextual advertising on the Web and the more I hear about the virtues of tags on general unstructured content, the more I see their failures in accomplishing their stated goals.
Let me soften this comment a bit by saying that these technologies do well on easy contextualization, or what I’d prefer to call word matching. For example, type in “kitesurfing” into the Google search engine and all of the AdWords ads that appear on the right column will use that word or be about kitesurfing. Pretty good one might say. Actually, these are excellent results because they’re all matching the one word selected to the one word paid for by the advertiser and the match is perfect.
Now go to a site that’s using AdSense, effectively the same technology, but now where some intelligence has to be applied to truly understand the content and context, and you’ll quickly find that the service doesn’t live up to the hype. It misinterprets or misses the essence of the context and content and simply tries to match on relevant key words in the text. Since there’s only one or two words to match on the ad side, it’s not likely that the hits will be very good.
Now tags are a different story. These are human generated, and any one can place any tag on any piece of content and have it appear in del.icio.us as a relevant bookmark. Or using a service like Flickr (pictures) or Consummating (dating), any one can tag (basically creating attributes of the content) their own content and place it for the world to access. Obvious problems here are; (1) that not all people will abide by the same rules for how they determine tag relevance, (2) there’s no normalization or ontology for the tags, so someone tagging an item with “bicycle” will not have this referenced to someone tagging an item with “bike”, (3) spamming tags is easy, simply take any message you want to spam and place every known tag to mankind ;-) As well, should any one determine a way to normalizing tagging, the challenge in reconciling all of the bad tags will not be a consistent exercise given that each person applied a different methodology for tagging their content.
I once heard an excellent analogy to this context and relevance issue. When looking for a song, knowing a note from that song won’t help you find it, you need to provide the melody. Similarly, finding context and relevance isn’t done by one word being matched, but rather by understanding the various meanings of several words, sentences or paragraphs.
All this to say, that so far all contextualization solutions are trying to take the easy way out and their use is marginal at best. This isn’t to say that Google’s AdSense program isn’t making bucket fulls of money, but it’s pure hype and with the low conversion rates they’re getting, further worsened by the amount of click-fraud being propogated, there’s not much difference between their so-called contextual ads and sheer chance of a relevant word match.
Progress still needs to be made here and already I’m seeing some technologies that are promising, but be careful not to be caught in the hype of context, ’cause today’s stuff is still far far from reaching that nirvana. In the words of “Chuck D” from Public Enemy, “Don’t believe the hype!”.