Article Preview
Top1. Introduction
The Web has become a popular venue to advertise. The online revenues have been steadily increasing at a staggering rate of 20% each year. The fact that total online revenues worldwide has risen to the tune of US $117 billion (statista.com, 2013) exemplify the significance of online advertising. A portion of this revenue comes from contextual advertising (Chakrabarti et al., 2008. Contextual advertising (CA) is essentially a mode of targeted advertising where the advertisement (ad) shown to the user is relevant to the webpage’s content. For instance, if a user is browsing a webpage about pizza, then the ads shown on the webpage may be of local pizza vendors. Webpage owners provide space on the webpage at primary locations to display such content-related ads. Users click on such ads and get directed to the ad webpage. Doing so not only brings revenues to the webpage owner but increases user experience as well, thereby giving rise to a win-win situation for both of the parties involved (Broder et al., 2007). In the current form of contextual advertising, when a webpage is being rendered in the user’s browser, the webpage requests ad(s) from an ad-providing entity known as ad-network. This request may contain a webpage URL and other associated information. The ad-network selects ads from its ad-repository by analyzing the request and responds by delivering content related ads. Therefore, it is of the utmost importance for the ad-network to select the most optimal content-related set of ads for the requesting webpage.
To match the ads with the webpage, an approach based on matching keywords between a webpage and that of the ad has been suggested (Ribeiro-Neto et al., 2005; Yih et al., 2006) in early work on CA. But this approach may lead to a bad selection of ads. A webpage about the damages of oil spills showing ad of an oil company is such an example. Such types of situations tarnish the reputations of both the webpage owner as well as the advertiser. To overcome these limitations, a semantic approach has been suggested (Broder et al., 2007, Anagnostopolulos et al., 2007, Armano et al., 2011; Lee et al., 2013). In this approach, a webpage and ads are classified into the nodes of a taxonomy. “Top matching” ads are retrieved to display on the webpage. The taxonomy used in CA is a hierarchical classification of a wide range of topics. If ads cannot be matched to the specific topic of the webpage, then ads are delivered on the generalized subject of the webpage. For example, if the subject of the webpage is swimming, then ads related to sportswear also serve the purpose of CA. Therefore, the semantic approach based ads are optimally related content-wise with the webpage.
For a semantic based CA system, a robust classification system resides at its core. The ad-network must classify the webpage based on its content’s subject into the nodes of the taxonomy of topics. The more accurately and specifically a webpage is classified, the better the ads will be retrieved from the ad-repository.