Evaluation and Comparison of a Novel Book Dataset for Aspect Based Sentiment Analysis
Author(s): Tamara Álvarez López, Milagros Fernández-Gavilanes, Enrique Costa-Montenegro, Patrice Bellot
Full text: submitted version
Abstract: Aspect-based sentiment analysis (ABSA) deals with extracting opinions at a fine-grained level in online texts, providing a very useful information for companies which want to know what people think about them or their products. Most of the systems developed on this field need a high amount of annotated data, nevertheless not many resources of this type can be found due to the high cost of preparation. In this paper we will introduce a new dataset, covering different subtasks, such us aspect extraction, category detection or sentiment analysis. It contains book reviews published in Amazon, which is a new domain in ABSA literature. We will introduce the dataset developed, the annotation process and its characteristics, as well as a comparison with other datasets, well-known in the field. This paper will focus on this comparison, addressing the different subtasks and analysing the performance and properties of each dataset.
Keywords: aspect-based sentiment analysis; book reviews; datasets; annotation; evaluation
Review 1 (by Pascal Hitzler)