Fernando Díez

In an era of increasing political polarization, its analysis becomes crucial for the understanding of democratic dynamics. This paper presents a comprehensive research on measuring political polarization on X (Twitter) during election cycles in Spain, from 2011 to 2019. A wide comparative analysis is performed on algorithms used to identify and measure polarization or controversy on microblogging platforms. This analysis is specifically tailored towards publications made by official political party accounts during pre-campaign, campaign, election day, and the week post-election. Guided by the findings of this comparative evaluation, we propose a novel algorithm better suited to capture polarization in the context of political events, which is validated with real data. As a consequence, our research contributes a significant advancement in the field of political science, social network analysis, and overall computational social science, by providing a realistic method to capture polarization from online political discourse.

Item Splitting is a context-aware recommendation technique based on Collaborative Filtering (CF), which groups and exploits ratings according to the contexts in which they were generated. It shows positive effects on recommendation accuracy in the presence of significant differences between the users' preferences from distinct contexts. To determine whether such differences are significant, in this paper we propose a novel impurity criterion based on the Fisher's exact test, which returns a score on the difference between ratings given to an item. Experimental results on a dataset of movie ratings show a lower rating prediction error with respect to other impurity criteria - in particular, related with time context signals - , letting us improve the recommendation performance of a state-of-the-art CF algorithm in an offline evaluation setting that simulates real-world conditions.

Item Splitting has been proposed as a technique for improving Collaborative Filtering (CF) by means of grouping and exploiting ratings according to the contexts in which they were generated. It shows positive effects on recommendation in the presence of significant differences between the users' preferences within distinct contexts. However, the additional user effort and specific system requirements needed to acquire contextual data may hamper the direct application of the above technique. In this paper we propose to split item sets using a number of time context representations derived from easy-to-collect rating timestamps. Initial results on standard datasets show that the proposed time contexts for item splitting let improve recommendation performance of a state-of-the-art CF algorithm in an offline evaluation setting simulating real-world conditions.

In the Social Web, a number of diverse recommendation approaches have been proposed to exploit the user generated contents available in the Web, such as rating, tagging, and social networking information. In general, these approaches naturally require the availability of a wide amount of these user preferences. This may represent an important limitation for real applications, and may be somewhat unnoticed in studies focusing on overall precision, in which a failure to produce recommendations gets blurred when averaging the obtained results or, even worse, is just not accounted for, as users with no recommendations are typically excluded from the performance calculations. In this article, we propose a coverage metric that uncovers and compensates for the incompleteness of performance evaluations based only on precision. We use this metric together with precision metrics in an empirical comparison of several social, collaborative filtering, and hybrid recommenders. The obtained results show that a better balance between precision and coverage can be achieved by combining social-based filtering (high accuracy, low coverage) and collaborative filtering (low accuracy, high coverage) recommendation techniques. We thus explore several hybrid recommendation approaches to balance this trade-off. In particular, we compare, on the one hand, techniques integrating collaborative and social information into a single model, and on the other, linear combinations of recommenders. For the last approach, we also propose a novel strategy to dynamically adjust the weight of each recommender on a user-basis, utilizing graph measures as indicators of the target user's connectedness and relevance in a social network.