Neural Budget Networks of Sensorial Data

This paper provides a neural network approach based on latent budget model to Sensorial Analysis of large data sets. Formally, consumer marker evaluations of products can be specified as preferences, budgets, or compositional data. For this type of data, unconditional as well as conditional latent budget analysis can be fruitfully considered. Main idea is to capture in few latent budget variables the consumer choices and the products sensorial properties. Main problem of neural network models is identification. A stabilizing algorithm within the Metropolis class is introduced. Applications of real data sets will be finally discussed