SentiTur: Building Linguistic Resources for Aspect-Based Sentiment Analysis in the Tourism Sector

Abstract

The use of linguistic resources beyond the scope of language studies, i.e. commercial purposes, has become commonplace since the availability of massive amounts of data and the development of tools to process them. An interesting focus on these materials is provided by Sentiment Analysis (SA) tools and methodologies, which attempt to identify the polarity or semantic orientation of a text, i.e., its positive, negative, or neutral value. Two main approaches have been made in this sense, one based on complex machine-learning algorithms and the other relying principally on lexical knowledge (Taboada et al., 2011). Lingmotif is an example of lexicon-based SA tool offering polarity classification and other related metrics, together with an analysis of the target segments evaluated (Moreno-Ortiz, 2017). Sentiment has been shown to be domain-specific to a large extent (Choi & Cardie, 2008) and it is therefore necessary to study and describe how sentiment is expressed not only in general language, but also in specialized domains. The availability of annotated, domain-specific corpora could greatly enhance the capacity of SA tools. Furthermore, the demand for a more fine-grained approach requires the identification of specific domain terminology, allowing the recognition of target terms associated with the polarity (Liu, 2012). Most available SA corpora are annotated at the document level, which allows systems to be trained to return the overall orientation of the text. However, more detail is necessary: what aspects exactly are being praised or criticized? This type SA is known as Aspect-Based Sentiment Analysis (ABSA), and attempts to extract more fined-grained knowledge. ABSA has attracted the attention of recent SemEval shared-tasks (Pontiki et al., 2015).