Identification of Client Profile Using Convolutional Neural Networks

In this work, a convolutional neural network is used to predict the interest of social networks users in certain product categories. The goal is to make a multi-class image classification to target social networks users as potential products consumers. In this paper, we compare the performance of several artificial neural network training algorithms using adaptive learning: stochastic gradient descent, adaptive gradient descent, adaptive moment estimation and its version based on infinity norm and root mean square prop. The comparison of the training algorithms shows that the algorithm based on adaptive moment estimation is the most appropriate to predict user's interest and profile, achieving about 99% classification accuracy.