Fusion Local and Global Aspect-based Sentiment Analysis

What is it about?

Aspected-based Sentiment Analysis (ABSA) is a fine-grained sentiment task in the field of NLP, and an important research direction in text sentiment analysis. It requires the model to be able to automatically extract aspects and predict the polarity of all aspects. For example, given a restaurant review: "The dessert at this restaurant is delicious, but the service is poor," the ABSA task requires extracting the two aspects of "dessert" and "service" and correctly reasoning their polarity. The ABSA task is divided into two subtasks: APC and ATE. The APC task aims to predict the exact emotional polarity of different aspects in their context, rather than vaguely analyzing the overall emotional polarity at the sentence level or text level. In APC tasks, emotional polarity is generally divided into three categories: positive, negative, and neutral. The ATE task is a sequence tagging task whose purpose is to extract aspects from comments or tweets.

Featured Image

Photo by Buzz Andersen on Unsplash

Photo by Buzz Andersen on Unsplash

Why is it important?

our model proposed in this paper combines local aspect word feature extraction and global contextual semantic information extraction based on Bi-directional Long Short-Term Memory (BiLSTM), and after a multi-headed attention mechanism to enhance the local aspect word sentiment representation. Comparative experiments were conducted on the restaurant and laptop datasets of the SEMEVAL2014 evaluation task. The experimental results show that the model proposed in this paper achieves good classification results in the aspect-level sentiment analysis task of text reviews. The method provides a new idea for ABSA task development.