Advancing Recommendations in Event-based Social Networks with Twin Tower Neural Networks and Embedding Techniques

Recommender Systems have become a vital part of nearly every business environment, continuously running in the background and constantly learning from users' habits and behaviours over time. However, commercially deployed recommender systems often exhibit high computational complexity, making their reusability and re-engineering time-consuming and expensive. Event-Based Social Networks (EBSNs) are particularly significant as they facilitate social interactions and community building by recommending relevant events to users. However, EBSNs are prone to cold start problems and tend to recommend events based primarily on users' geospatial data, rather than also considering their own preferences. To address these challenges, this study utilises a Twin Tower Neural Network Architecture. By passing precomputed Node2vec and Sentence Transformer Embedding through each tower, the proposed model is able to yield events tailored to the user's tastes, thereby enhancing user engagement. The model achieved an accuracy of 83.6% and a recall of 0.839 on the publicly available MeetUp Dataset, which consists of events and user data specific to Nashville, Tennessee, utilising it for training, validating, and testing,