Mining Fine-grained Opinion Expressions with Shallow Parsing

Proceedings of Recent Advances in Natural Language Processing, pages 302–310, Hissar, Bulgaria, 7-13 September 2013. Mining Fine-grained Opinion Expressions with Shallow Parsing Sucheta Ghosh CLCS, Trinity College Dublin ghoshs@tcd.ie Sara Tonelli FBK, Trento, Italy satonelli@fbk.eu Richard Johansson University of Gothenberg, Sweden richard.johansson@svenska.gu.se Abstract Opinion analysis deals with public opin- ions and trends, but subjective language is highly ambiguous. In this paper, we follow a simple data-driven technique to learn fine-grained opinions. We select an intersection set of Wall Street Jour- nal documents that is included both in the Penn Discourse Tree Bank (PDTB) and in the Multi-Perspective Question Answer- ing (MPQA) corpus. This is done in or- der to explore the usefulness of discourse- level structure to facilitate the extraction of fine-grained opinion expressions. Here we perform shallow parsing of MPQA ex- pressions with connective based discourse structure, and then also with Named Enti- ties (NE) and some syntax features using conditional random fields; the latter fea- ture set is basically a collection of NEs and a bundle of features that is proved to be useful in a shallow discourse parsing task. We found that both of the feature-sets are useful to improve our baseline at different levels of this fine-grained opinion expres- sion mining task. 1 Introduction The explosion of data in all forms from blogs, online forums, Facebook, Twitter and other so- cial media channels has given an opportunity of unprecedented reach to publicly sharing thoughts on events, products and services. However, there are some open issues related to this research area, commonly known as Opinion Mining, which can be summarized as follows: (1) Opinions are po- tentially ambiguous, and (2) Contextual interpre- tation of polarity is hard to achieve. Subsidiary important problem is the non-availability of large corpora with good annotation quality. Fine-grained opinion analysis is a different task from the coarse-grained one (e.g. document level analysis), in that it classifies opinion phrases, chunks or expressions from a given text. In this work, we perform fine-grained analysis by focusing on higher-level linguistic structure like discourse, without rich linguistic or knowledge- intensive features, to classify subjective opinion expressions using the Multi-Perspective Question Answering corpus (MPQA) scheme Wiebe et al. (2005). We perform two different experiments sets. We first exploit gold features based on shallow dis- course structure1 to classify fine-grained opinion expressions. In a second experiment, we use some syntax based features, those are found useful on a shallow discourse structure classification task, along with the named entities. Both of the experi- ments are found to be useful at different levels of fine-grained opinion expression mining. We use conditional random fields for this entire shallow parsing task. A set of documents from the Wall Street Journal (WSJ) corpus Marcus et al. (1993) annotated both in the Penn Discourse Treebank Prasad et al. (2008) and MPQA corpus is used. We also take advantage of the availability of sev- eral robust natural language processing tools pre- trained on WSJ data. 2 Related Work Fine-grained sentiment analysis methods have been developed …

Content truncated to 500 words