In this paper, we present a new Web Usage Mining method based on a semiotic analysis of the perceptual navigation styles, considering the user experience as a key element of his response to a given sollicitation. This study led us to define new semiotic features, which introduce an additional level to the classical semantic-based annotation approach. We also proposed a recurrent neural model to learn to treat these features, and explained why this model is well suited to process this kind of data.