The concept is very simple: Analyze the content of a webpage and extract from within the text, the keywords against which advertisers have bid. Select from the plethora of waiting advertisers, the most commercially relevant advertisements to display alongside the text. Place these alongside the content and then wait for the clicks to come in.
Where and when contextual advertising works, it is extremely accurate and effective. This is certainly true when the search is very specific and, therefore, the page content is also likely to be only about that particular context. For example, if you want to buy a television, enter the model name and number and the ads will offer an extremely attractive range of prices and vendors.
Conversely, where contextual advertising fails, it is so off the mark that it is a wasted impression. The root of the problem lies with the analysis of the page and the simplistic reliance upon the ability to use a single keyword, or combinations of keywords, to derive the context of an entire page. If only it were so easy!
Such a crude model could never hope to cope with the semantic complexity of a language. And the evidence of failure is all around us. To illustrate, when you read a review of the movie ‘Troy,’ the ads on the page persuade you to buy Trojan condoms. You read a report of a street stabbing on an online news site, and the onscreen ad suggests that you buy knives and cutlery. In the first case, the incorrect assignments of ads lead to wasted ad impressions and lost sales opportunities. In the second case, an over-simple keyword match made an appalling social error, with potential damage to the brands disclosed in the ads.
The reason for this is the general Orwellesque approach that some words are more important than others. Simply, bidded keywords have value, other words on the page do not. Unfortunately, in the garbage that is discarded are the words that help to derive the context of the page as a whole. It is this context which should be used to reduce the likelihood of the earlier examples occurring and to enhance the performance of the campaigns.
Also within the garbage are found the words that help to identify the actual sense of the terms being used. It is an often-ignored point that most words in an English language dictionary have an average of 2.4 alternative senses or meanings. Added to this are terms used as brand names, people’s names, domain names, etc., which increase the average to more than four alternative meanings for most words. If you rely on there only being one sense of any word, the likelihood of ad misplacement increases exponentially.
It should be accepted that not all of these issues are necessarily the fault of the targeting technology. Advertisers must accept a significant amount of the blame due to poor keyword selection. In selecting generic terms or failing to appreciate alternative meanings of a given term, the advertisers could well be wasting their own money and reducing the effectiveness of their own campaigns. That said, the fact that contextual advertising as a technology has failed to evolve in order to counter this issue is justifiably the fault of the current providers.
Quite obviously, to distinguish the different senses in such examples as the above, we need to identify exactly which associated words in the sites and ads relate to the required meaning. Processing the data using a simple statistical metric – such as looking for the most frequently occurring words – has not worked. (If it had, the above errors would never have happened!) Identifying relevant context is something the human brain does instantly and intuitively. Capturing this intuition, and turning it into a workable procedure, is a very different matter. All ambiguities have to be anticipated. It is not, after all, just a matter of a two-way choice. There are half-a-dozen commonly used senses of the word “bridge” in English, and if a site or ad wants to focus on one of these, interference from the others has to be eliminated.
A radically different approach is clearly needed, and current research is focused on finding alternatives. The door has been left open for an increasing number of providers, including our own, to develop solutions based upon the semantic targeting of ads. This technology aims to analyze not just bidded keywords, but all of the words on a webpage in order to deliver a more structured categorization of the page content and identify the specific sense of a word in true context.
These linguistics-based solutions analyze the lexical content of pages and link them with a more granular level of advertising. In addition they will be able to gauge, from the language used, the potentially objectionable nature of a page as well as the positive or negative sentiment of it. Through semantic targeting it is therefore not only possible for an advertiser to suggest in which context they would like their advertisement to be placed, but for that ad to only be placed where the content is discussing their product in a positive light.
The aim of semantic targeting products is to bring a new era of coherence to online contextual advertising. Building on the successes of the contextual advertising, they will bring about even greater benefits to advertisers and publishers alike.
By Sacha Carton