<p>Influence detection within social networks is a critical challenge for recommendation systems and marketing strategies. Classical machine learning often depends on user activity and engagement metrics as input features, which are usually crafted manually. However, these methods do not capture intricate relationships that are essential in social networks. Graph Neural Networks (GNNs) provide an alternative approach to deal with this problem by learning directly from the structure of the social graph. They overcome the shortcomings of manual encodings through automating the learning of interaction patterns that are typically difficult to define algorithmically. In our study, we analyze both classical and GNN models for influence classification, highlighting the efficiency of the graph-based model for understanding user behavior and inter-network relationships. Our results show that while classical methods can yield reasonably satisfying outputs with careful feature engineering, GNNs offer a more robust solution for representing influence in complex social systems.</p>