OPFI: A Tool for Opinion Finding in Polish

OPFI: A Tool for Opinion Finding in Polish Aleksander Wawer Institute of Computer Science, Polish Academy of Science ul. Jana Kazimierza 5, 01-238 Warszawa, Poland axw@ipipan.waw.pl Abstract The paper contains a description of OPFI: Opinion Finder for the Polish Language (Wawer, 2016), a freely available tool for opinion target extraction. The goal of the tool is opinion finding: a task of identifying tuples composed of sentiment (positive or negative) and its target (about what or whom is the sentiment expressed). OPFI is not dependent on any particular method of sentiment identification and provides a built-in sentiment dictionary as a convenient option. Technically, it contains implementations of three different modes of opinion tuple generation: one hybrid based on dependency parsing and CRF, the second based on shallow parsing and the third on deep learning, namely GRU neural network. The paper also contains a description of related language resources: two annotated treebanks and one set of tweets. Keywords: sentiment analysis, opinion targets, opinion tuples, opinion mining The paper is a description of OPFI: Opinion Finder for the Polish Language (Wawer, 2016), a freely available tool for opinion target extraction. It is organized as follows: Sec- tion 1. is an introduction to the problem of opinion tar- get extraction. Section sec:Overview contains a brief de- scription of opinion target identification methods, bundled in OPFI. Section 4. is a technical overview of OPFI and a discussion of input formats and involved tools. Section 5. provides a brief evaluation of selected OPFI algorithms and models. Finally, Section 6. concludes the paper. 1. Introduction Opinion extraction is about identifying and categorizing opinions expressed in a piece of text (as for example, senti- ments expressed in a word, phrase, sentence or document). Opinions are expressed about topics or objects, as for exam- ple positive or negative opinions about a product. Typically, opinion extraction consists of multiple sub-tasks, such as identifying opinions and their polarities, often called senti- ment analysis, followed by finding their targets (topics or objects of opinions). The goal of the latter sub-task, called opinion target extrac- tion, consists of identifying words towards which an opin- ion (sentiment) is expressed. Their character, meaning and syntactic role can vary and is influenced by corpus type. In the domain of product reviews, opinion targets are usually aspect terms, sometimes called attributes, denoting proper- ties of evaluated entity (eg.: This camera has great zoom but poor battery). It is also the case that it is entity itself that is the target of an opinion (eg.: I hate this camera). In other types of texts, opinion targets are more varied. They might be syntactically expressed not only as nouns and noun phrases, but as verbs referring to various states and activities (eg.: I do not like swimming). The formulation of opinion target extraction problem that we follow is similar to (Qiu et al., 2011), where authors also consider multiple types of opinion targets, including enti- ties and their aspects. To extract opinion targets, they apply two simple dependency patterns …

Content truncated to 500 words