Fecha: 24 de febrero de 2015

Ponente: Tamara Martín Wanton (UNED)

Lugar de celebración: Sala 1.03, ETSI Informática, UNED (mapa)


In this talk, I will present our ongoing work in Alert Detection, a
challenging task in Online Reputation Managament that consists in to
detect whether a topic should be an alert for the reputation of an
entity. We design topic-based features to distinguish between
non-important topics and alerts. Using Exploratory Factor Analysis
(EFA) we identify complex interrelationships among features and group
them as unified concepts called factors. These factors for the task of
alert detection can be interpreted as: /Novelty/ (features that detect
how much new information is present in topics compared with past
data), /Polarity/ (features related with the sentiment expressed in
topics); /Cohesiveness/ (if a topic is around a central theme); /User
Authority/ (features related with information about users).