Next-item recommendation within a short session using the combined features of horizontal and vertical convolutional neural network

Session-based recommendation systems are designed to offer recommendations to users based on their current browsing session rather than relying on their entire historical behavior. In many real-world scenarios, user profiles and historical behaviors are not readily available, which makes it challenging to provide accurate recommendations. However, most recommender systems only consider the user’s long-term profile and static preferences while ignoring their dynamic preferences, resulting in unreliable recommendations. The existing traditional and deep learning-based methods generate recommendations based on session data, such as clicks are not able to capture sequential patterns, contextual information, and dynamic preferences altogether. To address these issues, a deep learning-based model, i.e., horizontal vertical convolutional neural network (HV-CNN) has been proposed, which uses the combination of horizontal and vertical convolutional features to recommend the next item for a given sequence of items in the current ongoing session. The session clicks present in the dataset have been embedded using Word2Vec embedding technique before providing it to the proposed HV-CNN model. Although predicting the next item within a session is challenging due to the limited contextual information available, the proposed model outperforms state-of-the-art methods on the publicly available 30Music dataset.