I am often asked, as a linguist, to comment on the notions of relevance, coherence, and accuracy, which are commonly cited as desiderata in contextual advertising. Indeed, a widespread criticism of this domain is that, notwithstanding the emphasis on context, pages still manifest huge numbers of irrelevant results. So 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 organizations will have different views of what is relevant in a particular document; even within an organization, what may seem relevant to one person may not seem relevant to another. And even for an individual, what may seem relevant today may not seem so tomorrow. Any ad management system which does not allow for this inherent relativity is bound to fail.
Relevance is also slippery because most documents are multi-thematic. While it is of course possible to find documents which are 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, and so on contain more than one theme – often three or four. This can be sensed immediately from the headlines, as this selection from CNN illustrates.