Bilingual and cross domain politics analysis

Abstract

Opinion mining on Twitter recently attracted research in terest in politics using Information Retrieval (IR) and Natural Lan guage Processing (NLP). However, getting domain-specific annotated data still remains a costly manual step. In addition, the amount and quality of these annotation may be critical regarding the performance of machine learning (ML) based systems. An alternative solution is to use cross-language and cross-domain sets to simulate training data. This paper describe a ML approach to automatically annotate Spanish tweets dealing with the online-reputation of politicians. Our main finding is that a simple statistical NLP classifier without in-domain training can provide as reliable annotation as humans annotators and outperform more specific resources such as lexicon or in-domain data.