Aspect On: an Interactive Solution for Post-Editing the Aspect Extraction based on Online Learning

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4974–4981 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4974 Aspect On: an Interactive Solution for Post-Editing the Aspect Extraction based on Online Learning Mara Chinea-Rios, Marc Franco-Salvador, and Yassine Benajiba Symanto Research, N¨urnberg, Germany {mara.chinea, marc.franco, yassine.benajiba}@symanto.net Abstract The task of aspect extraction is an important component of aspect-based sentiment analysis. However, it usually requires an expensive human post-processing to ensure quality. In this work we introduce Aspect On, an interactive solution based on online learning that allows users to post-edit the aspect extraction with little effort. The Aspect On interface shows the aspects extracted by a neural model and, given a dataset, annotates its words with the corresponding aspects. Thanks to the online learning, Aspect On updates the model automatically and continuously improves the quality of the aspects displayed to the user. Experimental results show that Aspect On dramatically reduces the number of user clicks and effort required to post-edit the aspects extracted by the model. Keywords: aspect extraction, online learning, post-editing, annotation, user interface 1. Introduction Aspect extraction is the task of automatically extracting opinion targets in text (Liu, 2012). For instance, from the service was excellent, it aims to extract service. This task is a key component of aspect-based sentiment analysis (Hu and Liu, 2004), where the objective is to extract the senti- ment polarity associated with each aspect. The task of aspect extraction consists of two principal steps: i) extract all the aspect terms from a dataset, e.g. cleaner, laundry, coffee, pasta; ii) group the aspect terms into cat- egories that represent the same aspect, e.g. [cleaner, laun- dry] and [potatoes, pasta] for the cleaning and food aspects, respectively. In addition, there is a post-processing step that might be done to ensure the quality of the aspects. It usually comprises a manual revision and edition of the extracted as- pects and of the aspect terms annotated in a dataset. This step is specially important in production environments or for creating high quality aspect-based gold standards (Ganu et al., 2009; McAuley et al., 2012). One of the most accessible ways to post-edit the output of a model is through a User Interface (UI) (Wang and Soong, 2009). This method improves efficiency and productivity by providing users with tools specifically designed for their task. Furthermore, UIs provided with guidelines reduce the expertise required by the annotators, which makes it easier to find qualified assets. In case we want to include an online learning to optimize the task, UIs then, become an even more important component. Throughout this paper, we use the definition of online learn- ing as a method in which data becomes available in a se- quential order and is used to update our best predictor for future data at each step (Anderson, 2008). It is specially used in scenarios where it is computationally infeasible to train over the entire dataset or when the model continuously needs to …

Content truncated to 500 words