Automatic classification and PLS-PM modeling for profiling reputation of corporate entities on twitter

Abstract

In this paper, we address the task of detecting the reputation alert in social media updates, that is, deciding whether a new-coming content has strong and immediate implications for the reputation of a given entity. This content is also submitted to a standard typology of reputation dimensions that consists in a broad classification of the aspects of an under public audience company. Reputation manager needs a realtime database and method to report what is happening right now to his brand. However, typical Natural Language Processing (NLP) approaches to these tasks require external resources and show non-relational modeling. We propose a fast supervised approach for extracting textual features, which we use to train simple statistical reputation classifiers. These classifiers outputs are used in a Partial Least Squares Path Modeling (PLS-PM) system to model the reputation. Experiments on the RepLab 2013 and 2014 collections show that our approaches perform as well as the state-of-the-art more complex methods.