MultiBooked: A Corpus of Basque and Catalan Hotel Reviews Annotated for Aspect-level Sentiment Classification

MultiBooked: A Corpus of Basque and Catalan Hotel Reviews Annotated for Aspect-level Sentiment Classification. Jeremy Barnes, Patrik Lambert, Toni Badia Pompeu Fabra University, Barcelona, Spain {firstname.lastname}@upf.edu Abstract While sentiment analysis has become an established field in the NLP community, research into languages other than English has been hindered by the lack of resources. Although much research in multi-lingual and cross-lingual sentiment analysis has focused on unsupervised or semi-supervised approaches, these still require a large number of resources and do not reach the performance of supervised approaches. With this in mind, we introduce two datasets for supervised aspect-level sentiment analysis in Basque and Catalan, both of which are under-resourced languages. We provide high-quality annotations and benchmarks with the hope that they will be useful to the growing community of researchers working on these languages. Keywords: basque, catalan, sentiment analysis, aspect-level, under-resourced, opinion mining, cross-lingual 1. Introduction Sentiment analysis has become an established field with a number of subfields (aspect-level sentiment analysis, so- cial media sentiment analysis, cross-lingual sentiment anal- ysis), all of which require some kind of annotated resource, either to train a machine-learning based classifier or to test the performance of proposed approaches. Although much research into multi-lingual and cross- lingual sentiment analysis has focused on unsupervised or semi-supervised approaches (A.R. et al., 2012; Perez-Rosas et al., 2012; Gao et al., 2015), these techniques still require certain resources (linked wordnets, seed lexicon) and do not generally reach the performance of supervised approaches. In English the state-of-the-art for binary sentiment analysis often reaches nearly 90 percent accuracy (Tai et al., 2015; Kim, 2014; Irsoy and Cardie, 2014), but for other languages there is a marked drop in accuracy. This is mainly due to the lack of annotations and resources in these languages. This is especially true of corpora annotated at aspect-level. Unlike document- or tweet-level annotation, aspect-level annotation requires a large amount of effort from the an- notators, which further reduces the likelihood of finding an aspect-level sentiment corpus in under-resourced lan- guages. We are, however, aware of one corpus annotated for aspects in German (Klinger and Cimiano, 2014), al- though German is not a particularly low-resource language. The movement towards multi-lingual datasets for sentiment analysis is important because many languages offer differ- ent challenges, such as complex morphology or highly pro- ductive word formation, which can not be overcome by fo- cusing only on English data. The novelty of this work lies in creating corpora which cover both Basque and Catalan languages and are anno- tated in such a way that they are compatible with simi- larly compiled corpora available in a number of languages (Agerri et al., 2013). This allows for further research into cross-lingual sentiment analysis, as well as introduc- ing the first resource for aspect-level sentiment analysis in Catalan and Basque. The corpus is available at http: //hdl.handle.net/10230/33928 or https:// jbarnesspain.github.io/resources/. 2. Related Work In English there are many datasets available for document- and sentence-level sentiment analysis across different do- mains and at different levels of annotation …

Content truncated to 500 words