Integrated Sentiment Analysis and Graph Neural Networks for Understanding Consumer Behavior

Customer sentiment analysis is a critical determinant of opinion on customer needs and service delivery across different sectors. Previous studies have applied such methods of traditional machine learning to this end, but often they fail to provide a rather accurate representation of human emotions and relationships expressed in text. While capable of yielding fairly good results in terms of accuracy and recall, as well as in comprehending complex dependencies in large and constantly evolving data sets, these models also have certain weaknesses. To overcome these issues, this paper presents a new sentiment analysis framework using Graph Neural Networks (GNNs). As GNNs are more efficient in handling high dependency and relational data, something which makes them most appropriate to be employed in analysis of customer interactions. The proposed method preprocesses the textual data using method such as tokenization, noise removal and word embedding (BERT) subsequently followed by the sentiment classification using GNN. The novelty lies because GNN can model data graph structures and obviously complicated relations between words and phrases unlike traditional models. It was implemented on a number of customer reviews and displayed higher accuracy (99%). From these results, it can therefore be seen that GNN has many advantages over the traditional approaches and can be deemed a strong solution for the practical sentiment analysis use cases.