Search

We perform a temporal analysis of the Twitter stream to investigate the evolution of unique events based on the burst of popularity of associated hashtags. We derive a classification of events according to the different patterns corresponding to the peak of the volume of exchanged message and to how these events propagate on any social network with the same characteristics as Twitter. We first provide a precise definition of unique events and correlate them to hashtags. With reference to a specific interval of time, the most popular - with respect to number of tweets- hashtags are then detected using the Seasonal Hybrid ESD (S-H-ESD) technique introduced by Twitter. After identifying the unique hashtags among the 1000 most popular, we have identified, through an unsupervised Machine Learning algorithm applied to the historical temporal series of hashtags limited around the maximum peak, the temporal patterns (clusters) of the events. Finally, using the Twitter features, for each cluster, we have studied both the process at the origin of the event and how they evolve over the network.

In this paper, we study how to detect the most influential users in the microblogging social network platform Twitter and their evolution over time. To this aim, we consider the Dynamic Retweet Graph (DRG) proposed in Amati et al. (2016) and partially analyzed in Amati et al. (IADIS Int J Comput Sci Inform Syst, 11(2) 2016), Amati et al. (2016). The model of the evolution of the Twitter social network is based here on the retweet relationship. In a DRGs, the last time a tweet has been retweeted we delete all the edges representing this tweet. In this way we model the decay of tweet life in the social platform. To detect the influential users, we consider the central nodes in the network with respect to the following centrality measures: degree, closeness, betweenness and PageRank-centrality. These measures have been widely studied in the static case and we analyze them on the sequence of DRG temporal graphs with special regard to the distribution of the $75\%$ most central nodes. We derive the following results: (a) in all cases, applying the closeness measure results into many nodes with high centrality, so it is useless to detect influential users; (b) for all other measures, almost all nodes have null or very low centrality and (c) the number of vertices with significant centrality are often the same; (d) the above observations hold also for the cumulative retweet graph and, (e) central nodes in the sequence of DRG temporal graphs have high centrality in cumulative graph.

We propose a highly efficient and effective algorithm to estimate metrics on very large graphs that are based on the neighborhood function: examples of such metrics are the (effective) diameter and (effective) radius or the average distance. In order to efficiently provide good approximations to the size of the neighborhood set of any node, we refer to the MinHash Signatures approach to derive compressed representations of large sparse datasets but preserving similarity. The technique introduced here, named MinHash Signature Estimation (MHSE), exploits the similarity between signatures to estimate the size of the neighborhood function. We show that MHSE is as effective as HyperANF, which is considered the state of art approach for the estimation of the effective diameter of a very large graph. Indeed, performing both parametric (t-test) and non-parametric (Wilcoxon) statistical tests on residuals for average distance, effective diameter and number of connected pairs, the p-values show that MHSE tends to be more statistically significant than HyperANF. On the other hand, we show that MHSE is a very simple and easily distributable algorithm. In addition, by the property of the signatures to preserve similarity between neighborhoods of nodes, the algorithm can be suitably applied to allow to search and estimate the overlapping size of the most similar neighborhood at different distances.

Topological and structural properties of social networks, like Twitter, is of a major importance in order to understand the nature of user activities, for example how information propagates or how to identify influencing accounts. A deeper analysis of these properties may have a crucial impact on the design of new applications and of existing ones.

The study of the topological properties of graphs derived from social network platforms has a great importance both from the social and from the information point of view; furthermore, it has a big impact on designing new applications and in improving already existing services. Surprisingly, the research community seems to have mainly focused its efforts just on studying the most intuitive and explicit graphs, such as the follower graph of the Twitter platform, or the Facebook friends' graph: consequently, a lot of valuable information is still hidden and it is waiting to be explored and exploited. In this paper we introduce a new type of graph modeling behavior of Twitter users: the mention graph. Then we show how to easily build instances of this graphs starting from the Twitter stream, and we report the results of an experimentation aimed to compare the proposed graph with other graphs already analyzed in the literature, by using some standard social network analysis metrics.

In this paper we address the problem of combining a logic Λ with nonmonotonic modal logic. In particular we study the intuitionistic case. We start from a formal analysis of the notion of intuitionistic consistency via the sequent calculus. The epistemic operator M is interpreted as the consistency operator of intuitionistic logic by introducing intuitionistic stable sets. On the basis of a bimodal structure we also provide a semantics for intuitionistic stable sets.

We present tableau systems and sequent calculi for the intuitionistic analoguesIK, ID, IT, IKB, IKDB, IB, IK4, IKD4, IS4, IKB4, IK5, IKD5, IK45, IKD45 andIS5 of the normal classical modal logics. We provide soundness and completeness theorems with respect to the models of intuitionistic logic enriched by a modal accessibility relation, as proposed by G. Fischer Servi. We then show the disjunction property forIK, ID, IT, IKB, IKDB, IB, IK4, IKD4, IS4, IKB4, IK5, IK45 andIS5. We also investigate the relationship of these logics with some other intuitionistic modal logics proposed in the literature.

In this paper, we introduce a supervised method for the generation of a dictionary of weighted opinion bearing terms from a collection of opinionated documents. We also describe how such a dictionary is used in the framework of an algorithm for opinion retrieval, that is for the problem of identifying the documents in a collection where some opinion is expressed with respect to a given query topic. Several experiments, performed on the TREC Blog collection, are reported together with their results; in these experiments, the use of different combinations of DFR (Divergence from Randomness) probabilistic models to assign weights to terms in the dictionary and to documents is studied and evaluated. The results show the stability of the method and its practical utility. Moreover, we investigate the composition of the generated lexicons, mainly focusing on the presence of stop-words. Quite surprisingly, the best performing dictionaries show a predominant presence of stop-words. Finally, we study the effectiveness of the same approach to generate dictionaries of polarity-bearing terms: preliminary results are provided.

Due to the complexity of topical opinion retrieval systems, standard measures, such as MAP or precision, do not fully succeed in assessing their performances. In this paper we introduce an evaluation framework based on artificially defined opinion classifiers. Using a Monte Carlo sampling, we perturb a relevance ranking by the outcomes of these classifiers and analyse how the opinion retrieval performance changes. In this way it is possible to assess the performance of an approach to opinion mining from that of the overall system and to clarify how relevance and opinion are affected by each other.

We investigate the effectiveness of both the standard evaluation measures and the opinion component for topical opinion retrieval. We analyze how relevance is affected by opinions by perturbing relevance ranking by the outcomes of opinion-only classifiers built by Monte Carlo sampling. Topical opinion rankings are obtained by either re-ranking or filtering the documents of a first-pass retrieval of topic relevance. The proposed approach establishes the correlation between the accuracy and the precision of the classifier and the performance of the topical opinion retrieval. Among other results, it is possible to assess the effectiveness of the opinion component by comparing the effectiveness of the relevance baseline with the topical opinion ranking.

We study the evaluation of opinion retrieval systems. Opinion retrieval is a relatively new research area, nevertheless classical evaluation measures, those adopted for ad hoc retrieval, such as MAP, precision at 10 etc., were used to assess the quality of rankings. In this paper we investigate the effectiveness of these standard evaluation measures for topical opinion retrieval. In doing this we split the opinion dimension from the relevance one and use opinion classffiers, with varying accuracy, to analyse how opinion retrieval performance changes by perturbing the outcomes of the opinion classiffiers. Classiffiers could be studied in two modalities, that is either to re-rank or to filter out directly documents obtained through a first relevance retrieval. In this paper we formally outline both approaches, while for now focussing on the filtering process. The proposed approach aims to establish the correlation between the accuracy of the classiffiers and the performance of the topical opinion retrieval. In this way it will be possible to assess the effectiveness of the opinion component by comparing the effectiveness of the relevance baseline with that of the topical opinion.