Latent user models for online river information tailoring

Abstract

This paper explores Natural Language Generation techniques for online river information tailoring. To solve the problem of unknown users, we propose 'latent models', which relate typical visitors to river web pages, river data types, and river related activities. A hierarchy is used to integrate domain knowledge and latent user knowledge, and serves as the search space for content selection, which triggers user-oriented selection rules when they visit a page. Initial feedback received from user groups indicates that the latent models deserve further research efforts.