Christian Sánchez-Sánchez

Towards Automatic Detection of User Influence in Twitter by Means of Stylistic and Behavioral Features

Online communities are filled with comments of loyal readers or first-time viewers, that are constantly creating and sharing information at an unprecedented level, resulting in millions of messages containing opinions, ideas, needs and beliefs of Internet users. Therefore, businesses companies are very interested in finding influential users and encouraging them to create positive influence. Influential users represent users with the ability to influence individual’s attitudes in a desired way with relative frequency. We present an empirical analysis on influential users identification problem in Twitter. Our proposed approach considers that the influential level of users can be detected by considering its communication patterns, by means of particular writing style features as well as behavioral features. Performed experiments on more that 7000 users profiles, indicate that it is possible to automatically identify influential users among the members of a social networking community, and also it obtains competitive results against several state-of-the-art methods.

A Compact Representation for Cross-Domain Short Text Clustering

Nowadays, Twitter depicts a rich source of on-line reviews, ratings, recommendations, and other forms of opinion expressions. This scenario has created the compelling demand to develop innovative mechanisms to store, search, organize and analyze all this data automatically. Unfortunately, it is seldom available to have enough labeled data in Twitter, because of the cost of the process or due to the impossibility to obtain them, given the rapid growing and change of this kind of media. To avoid such limitations, unsupervised categorization strategies are employed. In this paper we face the problem of cross-domain short text clustering through a compact representation that allows us to avoid the problems that arise with the high dimensionality and sparseness of vocabulary. Our experiments, conducted on a cross-domain scenario using very short texts, indicate that the proposed representation allows to generate high quality groups, according to the value of Silhouette coefficient obtained.