DEEP CONVOLUTIONAL NEURAL NETWORKS for SENTIMENT ANALYSIS of CULTURAL HERITAGE

Abstract

The promotion of Cultural Heritage (CH) goods has become a major challenges over the last years. CH goods promote economic development, notably through cultural and creative industries and tourism. Thus, an effective planning of archaeological, cultural, artistic and architectural sites within the territory make CH goods easily accessible. A way of adding value to these services is making them capable of providing, using new technologies, a more immersive and stimulating fruition of information. In this light, an effective contribution can be provided by sentiment analysis. The sentiment related to a monument can be used for its evaluation considering that if it is positive, it influences its public image by increasing its value. This work introduces an approach to estimate the sentiment of Social Media pictures CH related. The sentiment of a picture is identified by an especially trained Deep Convolutional Neural Network (DCNN); aftewards, we compared the performance of three DCNNs: VGG16, ResNet and InceptionResNet. It is interesting to observe how these three different architectures are able to correctly evaluate the sentiment of an image referred to a ancient monument, historical buildings, archaeological sites, museum objects, and more. Our approach has been applied to a newly collected dataset of pictures from Instagram, which shows CH goods included in the UNESCO list of World Heritage properties.