Simple Unsupervised Similarity-Based Aspect Extraction
This addresses the need for efficient aspect extraction in sentiment analysis, offering a simpler alternative to computationally expensive deep learning methods, though it appears incremental as it builds on existing embedding techniques.
The paper tackled the problem of aspect extraction in sentiment analysis by proposing SUAEx, an unsupervised method based on word embedding similarity, which outperformed a state-of-the-art attention-based approach in experiments across three domains while being much faster.
In the context of sentiment analysis, there has been growing interest in performing a finer granularity analysis focusing on the specific aspects of the entities being evaluated. This is the goal of Aspect-Based Sentiment Analysis (ABSA) which basically involves two tasks: aspect extraction and polarity detection. The first task is responsible for discovering the aspects mentioned in the review text and the second task assigns a sentiment orientation (positive, negative, or neutral) to that aspect. Currently, the state-of-the-art in ABSA consists of the application of deep learning methods such as recurrent, convolutional and attention neural networks. The limitation of these techniques is that they require a lot of training data and are computationally expensive. In this paper, we propose a simple approach called SUAEx for aspect extraction. SUAEx is unsupervised and relies solely on the similarity of word embeddings. Experimental results on datasets from three different domains have shown that SUAEx achieves results that can outperform the state-of-the-art attention-based approach at a fraction of the time.