INTECIN   20395
INSTITUTO DE TECNOLOGIAS Y CIENCIAS DE LA INGENIERIA "HILARIO FERNANDEZ LONG"
Unidad Ejecutora - UE
congresos y reuniones científicas
Título:
Topic-based study of a Twitter political network
Autor/es:
J I ALVAREZ-HAMELIN; MARIANO G. BEIRO; T. MUSSI REYERO
Lugar:
Santiago
Reunión:
Conferencia; International School and Conference on Network Science; 2019
Institución organizadora:
Universidad del Desarrollo
Resumen:
Homophily and social influence are known to be the leading forces structuring social networks. Although different dynamical models for opinion diffusion have been theoretically studied, it is much harder to find traces of such evolution on real world data. The electoral period constitutes an excellent case study to investigate the consequences of social structure for democracy and political debate.In this study, we analyze a dataset of 54M tweets involving 300,000 users, captured throughout six months during the Argentinian presidential elections of 2015. These users were gathered from the followers of the main presidential candidates. We extract discussion topics from the tweet´s hashtags by building a co-occurrence weighted network in which a pair of hashtags is connected if they have been used together in the same tweet. From this network, we extract 4,000 topics using the OSLOM community detection algorithm [1], following the work in [2]. Figure 1 depicts the structure of the subgraph corresponding to a particular topic among the top-10 ones, based on the m-core decomposition used in [3]. The m-core decomposition is defined as the maximal subgraph whose edges participate in at least m triangles; and as k-core decomposition, it yields on a set of nested subgraphs that can discriminate between hierarchical and modular architectures.Regarding the topology of this instance of the Twitter network, we computed the homophily in followers, as it is done in [2]. The result in Figure 2 (left) shows, as expected, that the topical homophily (cosine similarity of the topics) is higher among followers than among random users, as also found in [2]. We also computed communities on the follower network using the OSLOM algorithm [1], and our results show that topical homophily is stronger when people belong to the same community of users (see Figure 2 right).Using the topics dimension defined from the hashtag network and decomposing the data (tweets of each user) into 15 days windows, we analyze the evolution of the alignment of the topics discussed by each user with those discussed by the candidates, along the time.