Platform for the discovery of newsworthy events in Twitter

• Autor(a) principal: Duarte, Fernando José Fradique
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/24806
• topic: Social Media; Twitter; Event Detection; Machine Learning; Dynamic Programming

The new communication paradigm established by Social Media, along with its growing popularity in recent years, have contributed to attract an increasing interest by several research fields. One such research field is the field of event detection in Social Media, whose relevance stems from its potential applicability in many diverse applications. One such application is the detection of newsworthy events in Social Media. The purpose of this work is therefore to implement a system to detect newsworthy events in Twitter. A similar system proposed in the literature is used as the base of this implementation. For this purpose a segmentation algorithm was implemented using a dynamic programming approach in order to split the tweets into segments. A weighting scheme that takes into account the burstiness, user support and newsworthiness of the segments was then used to rank these segments. Wikipedia was leveraged in order to derive this newsworthiness. The top K segments in this ranking were further processed and clustered into candidate events according to their similarity. These candidate events were then filtered by an SVM model trained on manually annotated data in order to retain only those related to real-world newsworthy events. The support infrastructure required by the system, namely regarding the precomputed values considered necessary to its operation was also implemented. The implemented system was tested with three months of data, representing a total of 4,770,636 tweets created in Portugal and mostly written in the Portuguese language. The precision obtained by the system was 76.9 % with a recall of 41.6%.