SMARTies: Sentiment Models for Arabic Target entities

Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 1002–1013, Valencia, Spain, April 3-7, 2017. c⃝2017 Association for Computational Linguistics SMARTies: Sentiment Models for Arabic Target entities Noura Farra and Kathleen McKeown Columbia University Department of Computer Science {noura,kathy}@cs.columbia.edu Abstract We consider entity-level sentiment anal- ysis in Arabic, a morphologically rich language with increasing resources. We present a system that is applied to com- plex posts written in response to Ara- bic newspaper articles. Our goal is to identify important entity “targets” within the post along with the polarity expressed about each target. We achieve signif- icant improvements over multiple base- lines, demonstrating that the use of spe- cific morphological representations im- proves the performance of identifying both important targets and their sentiment, and that the use of distributional semantic clus- ters further boosts performances for these representations, especially when richer linguistic resources are not available. 1 Introduction Target-specific sentiment analysis has recently be- come a popular problem in natural language pro- cessing. In interpreting social media posts, anal- ysis needs to include more than just whether peo- ple feel positively or negatively; it also needs to include what they like or dislike. The task of finding all targets within the data has been called “open-domain targeted sentiment” (Mitchell et al., 2013; Zhang et al., 2015). If we could suc- cessfully identify the targets of sentiment, it would be valuable for a number of applications includ- ing sentiment summarization, question answering, understanding public opinion during political con- flict, or assessing needs of populations during nat- ural disasters. In this paper, we address the open-domain tar- geted sentiment task. Input to our system con- sists of online posts, which can be comprised of one or multiple sentences, contain multiple enti- ties with different sentiment, and have different Figure 1: Online post with annotated target enti- ties and sentiment (green:pos, yellow:neg). domains. Our goal is to identify the important entities towards which opinions are expressed in the post; these can include any nominal or noun phrase, including events, or concepts, and they are not restricted to named entities as has been the case in some previous work. The only con- straint is that the entities need to be explicitly men- tioned in the text. Our work also differs from much work on targeted sentiment analysis in that posts are long, complex, with many annotated targets and a lack of punctuation that is characteristic of Arabic online language. Figure 1 shows an ex- ample post, where targets are either labeled posi- tive (green) if a positive opinion is expressed about them and negative (yellow) if a negative opinion is expressed. To identify targets and sentiment, we develop two sequence labeling models, a target-specific model and a sentiment-specific model. Our mod- els try to learn syntactic relations between enti- ties and opinion words, but they also make use of (1) Arabic morphology and (2) entity seman- tics. Our use of morphology allows us …

Content truncated to 500 words