Document modelling for customised information delivery

As the amount of information available to people multiplies at an increasing speed, it becomes ever more important to deliver information customised to users’ specific needs. Natural Language Generation systems coupled with user modeling techniques have been built to address this issue, to produce information that is relevant to the users . A common approach adopted by such systems is an approach based on planning, starting from a discourse (or communicative) goal, and planning the text to be presented to the users. However, these systems are not easy to build and difficult to change by domain experts. One of the problems is that it is hard to specify the plans employed, because they often require knowledge about writing, domain expertise, knowledge of computational linguistics and, finally, knowledge about how to obtain data from the underlying information sources. In this paper, we present our first step to address this problem.1