Awesome
The SentiAspect-pt corpus comprises 180 annotated product reviews at fine-grained opinion level. We manually annotated implicit and explicit aspects and hierarchically organized them for aspect-based sentiment analysis and opinion summarization applications in Portuguese.
<h2 align="left"> SentiAspect-pt Corpus </h2> <p align="justify"> Natural languages are very rich and allow us to express subjectivity in different ways. Consumers, for example, may use implicit or explicit aspects to refer to the same product property. e.g., the sentences “she got calls at the São Francisco river” and “working anywhere” have been employed in actual reviews to evaluate the (implicit) “signal” aspect of a smartphone. Therefore, not every opinion is directly expressed and not every aspect appears explicitly in the text. For example, "The camera is expensive", in this case, the evaluated aspect is “price”, but it is implicit, not being explicitly said in the sentence and, therefore, must be inferred from the context.</p> <h2 align="left"> Annotation Process</h2> <p align="justify"> We annotated 180 reviews according to implicit and explicit fine-grained opinions (aspect-based sentiment analysis). For the identification of implicit aspects, we also annotated the "clue terms" that indicated the aspects. For example, in "This camera is expensive", the aspect that was evaluated here is "price", but it is implicit. The term “expensive” is, therephore, the "clue term" to identify the aspect "price". The identification of explicit aspects has directly annotated. For example, "The history of the book is bad". In this case, “history” is an explicit aspect. In the last stage, the aspects were gruped according to their similar meanings. For example, the “cost”, “value”, “price” and “investment” aspects are aspects of the same group.</p>
