A Hierarchical DAG-based Neural Network Approach to Aspect-based Sentiment Analysis

The DNN-BSA (DAG based Neural Networks for Sentiment Analysis) technique, which is used in the context of sentiment analysis, focuses on reliably recognizing opinion aspects and the sentiment polarity that correlate to those opinion features. Both stand-alone and integrated approaches of the past have proven unsuccessful in their attempts to accurately capture the intrinsic link that exists between these two goals. In order to overcome this issue, the study that we have conducted suggests the use of an innovative dynamic DAG graph that explicitly reflects the relationship that exists between aspect extraction and sentiment identification. The DAG network that we use comprises both regular words and emotion labels as linked nodes. This is how we tackle the problem. This approach enables interaction between aspect words and emotion information to take place in a fluid manner. During the process of real-time prediction, the graph is dynamically changed after being initialized with many distinct types of dependencies. The performance of our model is superior to that of other state-of-the-art models, as demonstrated by numerous tests conducted on benchmark datasets. In addition, in-depth examination suggests that our model exhibits considerable gains when dealing with difficult scenarios that involve multiple-opinion aspects as well as circumstances in which there is no opinion element present. This demonstrates how resilient and successful our methodology is in dealing with difficult sentiment analysis assignments.