Rexha Andi, Mauro Dragoni, Marco Federici
ReUS: A Real-Time Unsupervised System For Monitoring Opinion Streams
Background An actual challenge within the sentimentanalysis research area is the extraction of polarity valuesassociated with specic aspects (or opinion targets)contained in user generated content. This task, calledaspect-based sentiment analysis bring new challengeslike the disambiguation of words' role within text andthe inference of correct polarity values based on thedomain in which a text occurs. The former requiresstrategies able to understand how each word is usedin a specic context in order to annotate it as aspector not. The latter need to be addressed with unsupervisedsolutions in order to make a system ecient forreal-time tasks and at the same time exible in order toadopt it in any domain without requiring the trainingof sentiment models. Finally, the deployment of suchsystem into real-world scenarios needs the developmentof usable solutions for accessing and analyzing data.Methods This paper presents the ReUS platform: a systemintegrating an unsupervised approach, based onopen information extraction strategies, for performingreal-time aspect-based sentiment analysis together withfacilities supporting decision makers in the analysis andvisualization of collected data.Results The ReUS platform has been validated from aquantitative and qualitative perspectives. First, the aspectextraction and polarity inference capabilities have been evaluated on three dataset used in likewise editionsof SemEval. Second, a user group has been invitedto judge the usability of the platform.Conclusion The developed platform demonstrated to besuitable for being used into real-world scenarios requiring(i) the capability of processing real-time opinionbaseddocuments streams and (ii) the availability ofusable facilities for analyzing and visualizing collecteddata. Examples of possible analysis and visualizationsincludes the presentation of lists ranking aspects bytheir importance of by their polarity values computedwithin the whole data repository. This kind of analysisenables, for instance, the discovery of product issues.