This article focuses on the use of linguistic “weak signals”—digital traces of intent—in social media as a tool of counterterrorism aimed at preventing lone-wolf attacks.
This 2014 article focuses on sentiment analysis and proposes a new model for analyzing user sentiments and opinions online. The authors problematize that most current topic methods used for sentiment analysis use probabilistic topic models that only model sentiment text, but do not consider the user, who expresses the sentiment, nor the item that the sentiment is expressed on. They argue that it is better to incorporate the user and item information into the topic model. The authors instead propose a Supervised User-Item based Topic model called the SUIT model. To demonstrate the advantages of their model, the authors conduct extensive experiments on two datasets, a review dataset and a microblog dataset, and compare their model against both supervised topic models and collaborative filtering methods.
This article will be of particular use to PVE researchers and practitioners interested in sentiment analysis and particularly those looking to expand beyond current models. There are a growing number of ways for people to express their sentiments online, including comments on news sites, microblogs, forums, and more. Sentiment analysis aims to analyze these expressed sentiments and opinions of users towards various topics, issues, and events, and better understanding these sentiments can provide PVE researchers with a more complete picture of a group, network or wider subset of a given population. The authors’ SUIT model is particularly advantageous as it can simultaneously utilize the textual topic and latent user-item factors. The model uses tensor outer product of text topic proportion vector, user latent factor and item latent factor to model the sentiment label generalization.