• by Sacha Carton , July 14, 2009

The goal of any ad campaign should be to make money for the advertiser. Be they measured in clicks, conversions, or brand awareness, it comes down to making money. So what is the role of relevance?

In comparing the performance metrics of an average contextual advertising campaign with a standard run of network display ad campaign, we see that relevance really does make a difference, often in hundreds of percentage points. Thus it appears that relevance can substantially improve the efficiency of an ad campaign.

It therefore stands to reason that the goal of any ad targeting system should be to display the most relevant ad in front of the user when they are reading relevant content. Catching them while they are 'in the zone' should mean that there is a higher chance of converting the interest into a click for a particular campaign. As result, the companies behind the contextual, behavioural or semantic targeting systems are all constantly attempting to tweak their technologies to enhance the user experience by delivering ads in 'true' context with the page content. So, just how is relevance to be improved?

The notion of relevance is slippery, because it always means 'relevant to an individual's needs', and these alter according to circumstances. Different clients will have different views of what is relevant on a given page; even within a single small group, what may be relevant to one person may not be relevant to another; and even for an individual, what may seem relevant today may not be so tomorrow. Absolute notions of relevance are therefore chimerical and any ad targeting system which does not allow for this inherent relativity is bound to fail.

Relevance is related to a user's sense of what a page is 'about'. But there is a complication: it is not always clear - especially in a lengthy passage of text - what the content is about. Moreover, our ability to sense what a document is about is affected by many variables, such as fatigue, stress, attention span, motivation, and knowledge of a field.

What tends to happen is that readers are swayed by the writer's perception of what a document is about, as seen through the headline or other heading. However, it is the nature of headlines to be selective so that sub-themes within a page may not figure there. This is the main reason why a simple keyword-based targeting system built on a simplified analysis of text for bidded keywords may appear somewhat alarmingly out of context. It is also the reason why people should not judge the context of an article solely by the headline. Using an example from our own work to illustrate my point, there was a news article about a tennis tournament champion that was given a classification of 'tennis' and 'race cars'. The latter seemed anomalous unless you read the whole page and found that halfway through the author included a section about the tennis player's husband, who was a successful race car driver.

Another reason I deem relevance as slippery is because most web pages or documents are multi-thematic. While it is of course possible to find some restricted to a single theme, the vast majority of everyday documents seen in the world's online news channels, travel guides, scientific reports, entertainment reviews, etc. contain more than one theme - often three or four.

But even if knowledge is present, other factors will interfere. Ask two people to identify all the themes in a document and they are unlikely to come up with the same set. For example, a webpage on outsourcing call centers will probably contain thematic elements relating to domains within economics, sociology, politics, and technology, as well as locational domains such as India. These elements will not be clearly distinguishable.

Given the above, it is easy to see why the concept of relevance in advertising terms has been extremely loose in its interpretation up until now. Advertising targeting systems seeking to apply relevance have extremely variable results, from the wildly inaccurate to the eerily precise. So just how is relevance applied?

In its most basic form, relevance is identified from an overall categorization at site level. A given travel site can leverage its theme and promote all possible campaigns relating to this subject. However, the classification of the content often takes place at site level and misses the mother lode of highly specific content and the commercial opportunities that lie within.

An improvement was made with the introduction of 'contextual' targeting. By identifying specific advertiser keywords within a page, corresponding ads can be placed alongside the content. Relevance is achieved but only in relation to the keywords and not the article as a whole. Without a systematic analysis of context of the page, the true sense of the keyword cannot be identified - and with more than 2.4 senses for each word in the English language, the ad's relevance to the page cannot be guaranteed.

Relevance is a stated aim for behavioral targeting systems. By identifying a reader's inferred traits as a result of a series of attributes such as what they read, what they click on and everything relating to their navigation around a site, behavioral targeting can get quite granular in delivering relevance. However, it is a different kind of relevance - relevance to the reader through their inferred behavior, not necessarily specific to the type of content that they are viewing. Unfortunately such relevance does come at a cost, with the methods of aggregation of such relevance forcing the reader to compromise on privacy as their web usage is tracked through cookies. Surely relevance can be found without having to resort to such tactics.

Genuine relevance can only ever be achieved with the identification of the granularity of content present on each and every webpage. Such a system requires the following:
•    Ability to understand the usage of a word based upon analysis of all of the words on a page
•    Identification of whether or not it would be prudent for an advertiser's brand to be associated with the type of content found on a webpage
•    Permits readers to maintain their inalienable rights to privacy

Does such a comprehensive system exist?

With the development and progress of semantic targeting technology, yes, such systems are now available. The study of meaning, semantics offers the opportunity for each nuance of a page to be identified and exploited for online advertising purposes.

Semantic targeting technology analyzes not only bidded keywords, but all of the words upon a webpage. In turn, this delivers a more detailed categorization of page content, as well as identification of the specific sense of a word in true context.

These linguistics-based solutions analyze the lexical content of pages and link it with a more granular level of advertising. And from the language used, they can gauge the sentiment of a page. It is therefore possible for advertisers to request which context they want their advertisements to be placed with and for them to only to be placed where their product/company is discussed in a positive light.
 
Relevance does matter and can make the difference between successful and mediocre online advertising campaigns. Through semantic targeting, we believe that true ad relevancy can be achieved and represents the next major development in online advertising, adding value to content by making every page a relevance-rich opportunity.