Session-aware Graph Neural Network-based Recommendations for Custom Advertisement Segment Generation
Tomasz Palczewski, Anirudh Rao · 2023
With the rapid rise of online services available to users, it is of paramount importance to keep users engaged by showcasing only the most relevant information to them. Recommendation engines play a pivotal role in this endeavor and find applications across a broad range of industries, such as e-commerce, advertising, marketing etc. In general, recommendation systems can be divided into two groups - 1) systems built on general interaction information and, 2) systems built on sequential interaction information. In this paper, we focus on the latter, since we believe that user tastes evolve over time and the system must leverage temporal information when making recommendations. As research in the field has progressed, there has been particular emphasis on ways to include different types of side information to enhance the contextual relevance of the recommendations. Our model, called Session-awaRe grAph Neutral Network-based Recommendation, or SRA-NN-Rec for brevity, builds upon this idea. The architecture of the model is inspired by two state-of-the-art sequence based approaches. The first model, called SR-GNN, is a session based model where the user identity is unknown and only access to current session of user interactions is available. The second model, called A-PGNN, is a session aware model where the user identity is known and one can leverage previous session information as a result. The experimental results on two publicly available datasets show that the SRA-NN-Rec performs reasonably well compared to existing baseline models and is an effective model for recommendations when sessionized information is available. We briefly discuss how this model could be used for custom advertisement segment generation.