Detecting and locating trending places using multimodal social network data

Abstract This paper presents a machine learning-based classifier for detecting points of interest through the combined use images and text from social networks. model exploits transfer learning capabilities neural network architecture CLIP (Contrastive Language-Image Pre-Training) in multimodal environments using image text. Different methodologies based on information are explored geolocation places detected. To this end, pre-trained models used classification their associated texts. The result is system that allows creating new synergies between texts order to detect geolocate trending has not been previously tagged by any other means, providing potentially relevant tasks such as cataloging specific types city tourism industry. experiments carried out reveal that, general, textual more accurate than visual cues setting.