Generalization Methods for In-Domain and Cross-Domain Opinion Holder Extraction

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 325–335, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Generalization Methods for In-Domain and Cross-Domain Opinion Holder Extraction Michael Wiegand and Dietrich Klakow Spoken Language Systems Saarland University D-66123 Saarbr¨ucken, Germany {Michael.Wiegand|Dietrich.Klakow}@lsv.uni-saarland.de Abstract In this paper, we compare three different generalization methods for in-domain and cross-domain opinion holder extraction be- ing simple unsupervised word clustering, an induction method inspired by distant supervision and the usage of lexical re- sources. The generalization methods are incorporated into diverse classifiers. We show that generalization causes significant improvements and that the impact of im- provement depends on the type of classifier and on how much training and test data dif- fer from each other. We also address the less common case of opinion holders being realized in patient position and suggest ap- proaches including a novel (linguistically- informed) extraction method how to detect those opinion holders without labeled train- ing data as standard datasets contain too few instances of this type. 1 Introduction Opinion holder extraction is one of the most im- portant subtasks in sentiment analysis. The ex- traction of sources of opinions is an essential com- ponent for complex real-life applications, such as opinion question answering systems or opin- ion summarization systems (Stoyanov and Cardie, 2011). Common approaches designed to extract opinion holders are based on data-driven methods, in particular supervised learning. In this paper, we examine the role of general- ization for opinion holder extraction in both in- domain and cross-domain classification. General- ization may not only help to compensate the avail- ability of labeled training data but also conciliate domain mismatches. In order to illustrate this, compare for instance (1) and (2). (1) Malaysia did not agree to such treatment of Al-Qaeda sol- diers as they were prisoners-of-war and should be accorded treatment as provided for under the Geneva Convention. (2) Japan wishes to build a $21 billion per year aerospace indus- try centered on commercial satellite development. Though both sentences contain an opinion holder, the lexical items vary considerably. How- ever, if the two sentences are compared on the ba- sis of some higher level patterns, some similari- ties become obvious. In both cases the opinion holder is an entity denoting a person and this en- tity is an agent1 of some predictive predicate (i.e. agree in (1) and wishes in (2)), more specifically, an expression that indicates that the agent utters a subjective statement. Generalization methods ide- ally capture these patterns, for instance, they may provide a domain-independent lexicon for those predicates. In some cases, even higher order fea- tures, such as certain syntactic constructions may vary throughout the different domains. In (1) and (2), the opinion holders are agents of a predictive predicate, whereas the opinion holder her daugh- ters in (3) is a patient2 of embarrasses. (3) Mrs. Bennet does what she can to get Jane and Bingley to- gether and embarrasses her daughters by doing …

Content truncated to 500 words