Geolocation Detection Approaches for User Discussion Analysis in Twitter

In this research, the authors consider methods for identifying geodata of users of social networks within user discussions. The knowledge of user geolocation data makes it possible to analyze the spread of discussion among users of different countries. Authors do not try to determine the exact geolocation, but rather the country where the users are located. The problem of getting country-level user location data lies in the fact that a high percentage of users do not state their location correctly, either mentioning it in humorous ways or even not stating it at all. There are various methods of obtaining data about the location of users. Among them, there are text-based methods, methods based on the analysis of the context, and methods based on the topology of the user graph. In this paper, we make a special emphasis on a method that allows to reveal geodata of users who specified their geodata incorrectly or did not specify it at all. In order to test our method, we use Twitter datasets. We propose several approaches to resolve the issues stated above. The paper highlights three approaches: the naive approach, the naive approach using natural language processing (NLP), and the graph approach, which is glossary-based and determines the number of outgoing connections. We have introduced twomeasures in order to evaluate the proposed approaches. Recall-GEO and Precision-GEO that are described throughout the paper. The accuracy of UserGraph method is finally evaluated using the metrics above.