Grammatical structures for word-level sentiment detection

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 667–676, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Grammatical structures for word-level sentiment detection Asad B. Sayeed MMCI Cluster of Excellence Saarland University 66123 Saarbr¨ucken, Germany asayeed@coli.uni-sb.de Jordan Boyd-Graber, Bryan Rusk, Amy Weinberg {iSchool / UMIACS, Dept. of CS, CASL} University of Maryland College Park, MD 20742 USA {jbg@umiacs,brusk@, aweinberg@casl}.umd.edu Abstract Existing work in fine-grained sentiment anal- ysis focuses on sentences and phrases but ig- nores the contribution of individual words and their grammatical connections. This is because of a lack of both (1) annotated data at the word level and (2) algorithms that can leverage syn- tactic information in a principled way. We ad- dress the first need by annotating articles from the information technology business press via crowdsourcing to provide training and testing data. To address the second need, we propose a suffix-tree data structure to represent syntac- tic relationships between opinion targets and words in a sentence that are opinion-bearing. We show that a factor graph derived from this data structure acquires these relationships with a small number of word-level features. We demonstrate that our supervised model per- forms better than baselines that ignore syntac- tic features and constraints. 1 Introduction The terms “sentiment analysis” and “opinion mining” cover a wide body of research on and development of systems that can automatically infer emotional states from text (after Pang and Lee (2008) we use the two names interchangeably). Sentiment analysis plays a large role in business, politics, and is itself a vibrant research area (Bollen et al., 2010). Effective sentiment analysis for texts such as newswire depends on the ability to extract who (source) is saying what (target). Fine-grained sen- timent analysis requires identifying the sources and targets directly relevant to sentiment bearing expres- sions (Ruppenhofer et al., 2008). For example, con- sider the following sentence from a major informa- tion technology (IT) business journal: Lloyd Hession, chief security officer at BT Radianz in New York, said that virtualiza- tion also opens up a slew of potential net- work access control issues. There are three entities in the sentence that have the capacity to express an opinion: Lloyd Hession, BT Radianz, and New York. These are potential opinion sources. There are also a number of mentioned con- cepts that could serve as the topic of an opinion in the sentence, or target. These include all the sources, but also “virtualization”, “network access control”, “network”, and so on. The challenging task is to discriminate between these mentions and choose the ones that are rele- vant to the user. Furthermore, such a system must also indicate the content of the opinion itself. This means that we are actually searching for all triples {source, target, opinion} in this sentence (Kim and Hovy, 2006) and throughout each document in the corpus. In this case, we want to identify that Lloyd Hession is the source of an opinion, “slew of network issues,” …

Content truncated to 500 words