Open Domain Targeted Sentiment

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1643–1654, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Open Domain Targeted Sentiment Margaret Mitchell Jacqueline Aguilar Theresa Wilson Benjamin Van Durme Human Language Technology Center of Excellence Johns Hopkins University Baltimore, MD 21218, USA {m.mitchell,jacqui.aguilar}@jhu.edu, Theresa.Wilson@oberlin.edu, vandurme@cs.jhu.edu Abstract We propose a novel approach to sentiment analysis for a low resource setting. The in- tuition behind this work is that sentiment expressed towards an entity, targeted senti- ment, may be viewed as a span of sentiment expressed across the entity. This represen- tation allows us to model sentiment detec- tion as a sequence tagging problem, jointly discovering people and organizations along with whether there is sentiment directed to- wards them. We compare performance in both Spanish and English on microblog data, using only a sentiment lexicon as an exter- nal resource. By leveraging linguistically- informed features within conditional random fields (CRFs) trained to minimize empiri- cal risk, our best models in Spanish signifi- cantly outperform a strong baseline, and reach around 90% accuracy on the combined task of named entity recognition and sentiment pre- diction. Our models in English, trained on a much smaller dataset, are not yet statistically significant against their baselines. 1 Introduction Sentiment analysis is a multi-faceted problem. De- termining when a positive or negative sentiment is being expressed is a large part of the challenge, but identifying other attributes, such as the target of the sentiment, is also crucial if the ultimate goal is to pinpoint and extract opinions. Consider the exam- ples below, all of which contain a positive sentiment: (1) So happy that Kentucky lost to Tennessee! (2) Kentucky versus Kansas I can hardly wait... (3) Kentucky is the best alley-oop throwing team since Sherman Douglas’ Syracuse squads!! The entities in these examples are college basket- ball teams, and the events referred to are games. In (1), although there is a positive sentiment, the tar- get of the sentiment is an event (Kentucky losing to Tennessee). However, from the positive sentiment toward this event, we can infer that the speaker has a negative sentiment toward Kentucky and a positive sentiment toward Tennessee. In (2), the positive sen- timent is toward a future event, but we are not given enough information to infer a sentiment toward the mentioned entities. In (3), Kentucky is the direct target of the positive sentiment. We can also in- fer a positive sentiment toward Douglas’s Syracuse teams, and even toward Douglas himself. These examples illustrate the importance of the target when interpreting sentiment in context. If we are looking for sentiments toward Kentucky, for ex- ample, we would want to identify (1) as negative, (2) as neutral (no sentiment) and (3) as positive. How- ever, if we are looking for sentiment toward Ten- nessee, we would want to identify (1) as positive, and (2) and (3) as neutral. The expression of these and other kinds of sen- timent can be understood as involving three …

Content truncated to 500 words