Location learning and prediction in Social Networks

Nowadays, the analysis of data from Online Social Networks (OSNs) is one of the main areas of interest for companies involved in data analysis. A particular type of OSNs are \ac{LBSN}s, which in addition to providing the normal functions of social networks add location-based services. This research aims to analyze the data of one of the most widely used LSBNs at the moment, namely Twitter, and calculate the entropy variation of the latter to identify any anomalies in different geographical areas. A careful analysis has been carried out on what is currently the panorama of data analysis made on the most used social networks at the moment. The result is the enormous difficulty in obtaining data from LSBNs due to the privacy restrictions imposed in recent years. In this project a code has been developed that allows to obtain in real time the geolocalized data coming from Twitter. Subsequently these data were analyzed and filtered to be subjected to different statistics, in this context was developed an interactive map in Python that allows to see the distribution of tweets in the different areas of the chosen geographical areas. The proposal is tested on a set of data collected by Twitter during a month in Madrid and compared with another set of data from Rome. Finally, an algorithm was applied to the data obtained for the calculation of entropy provided by the Department of Telematic Engineering of the University Carlos III, which allowed us to analyze the trend of the crowd during the period examined. Finally, this work aims to provide both the methods to obtain geolocalized data in Streaming from Twitter and those to analyze them carefully. Our goal is to find anomalies. The most obvious in a short trace in which nothing unusual happens is that the behavior on the weekend is different from the labour day, but the objective is to detect any other unusual behavior.