Profiling Users to Perform Contextual Advertising
Andrea Addis, Giuliano Armano, Eloisa Vargiu · 2009
Recommendation systems are typically aimed at proposing items to users. Nevertheless, another way of performing recommendations is viable, i.e. proposing users to domain specific web sites. Within this context, users require to be represented according to their preferences –given in terms of categories of interest. To better highlight the need for “recommending users”, let us recall that commercial web sites are typically involved with one or more domain-specific businesses. In this scenario, a system able to identify relevant categories can be useful to identify users that may become target for advertisement (for instance, a company that sells pet supplies, food and products is interested in identifying users that have the “Animals” category among their interests). The goal of this research is to develop agent-based referral systems for user profiling able to identify user interests. Given a taxonomy and a set of documents representing a user (i.e. selected by the user while surfing the web) the system is able to profile her/him in terms of the given categories. For the sake of simplicity, experiments have been performed using WordNet Domains as reference taxonomy and Wikipedia as document source. The underlying assumption is that, due to the general-purpose machine learning techniques adopted to implement the system, this capability is exportable to other –more specific– taxonomies.