Survey on Aspect Category Detection

Abstract

In recent years, aspect category detection has become popular due to the rapid growth in customer reviews data on e-commerce and other online platforms. Aspect Category Detection, a sub-task of Aspect-Based Sentiment Analysis, categorizes the reviews based on the features of a product such as a laptop’s display, or an aspect of an entity such as the restaurant’s ambiance. Various methods have been proposed to deal with such a problem. In this paper, we first introduce several datasets in the community that deal with this task and take a closer look at them by providing some exploratory analysis. Then, we review a number of representative methods for aspect category detection and classify them into two main groups: 1) supervised learning, and 2) unsupervised learning. Next, we discuss the strengths and weaknesses of different kinds of methods, which are expected to benefit both practical applications and future research. Finally, we discuss the challenges, open problems, and future research directions.

Machine-Learning-Based Approaches for Multi-Level Sentiment Analysis of Romanian Reviews

“…While it is crucial to obtain an understanding of user opinion through analysis at the document level, and decompose it further into a study at the sentence level, in reviews, users often make evaluations with respect to different aspects of a given product, where an aspect refers to a characteristic, behavior, or trait of a product or entity [9]. For instance, for mobile phones, aspect categories of interest to users, generally, are battery life, photo and video quality, sound, and performance.…”

“…As for ACD, which aims to identify the discussed aspect categories for a given sentence, most state-of-the-art approaches formalize the task as a supervised text classification problem where a generally small set of predefined, domain-specific aspect categories represent the classes [13]. Unsupervised formulations often involve two steps: first extracting candidate aspect terms (the ATE task), and then grouping or mapping these terms to corresponding aspect categories.…”

“…The experiments are conducted on a perfectly balanced corpus, consisting of 1000 movie reviews written in Romanian (500 positive and 500 negative), manually extracted from several blogs and websites. The documents have been manually labeled, based on the individual scores assigned by the user (in the range [1][2][3][4][5][6][7][8][9][10]). To identify document-level polarity, the authors experiment with random forest (RF), kNN, NB, and SVM, the maximum precision values obtained being 81.8% (using SentiWordnet) and 79.2% (using Bing queries).…”