As recommender systems shift from rating-based to interaction-based models, graph neural network-based collaborative filtering models are gaining popularity due to their powerful representation of user-item interactions. However, these models may not produce good item ranking since they focus on explicit preference predictions. Further, these models do not consider side information since they only capture latent feature information of user-item interactions. This study proposes an approach to overcome these two issues by employing preference relation in the graph neural network model for collaborative filtering. Using preference relation ensures the model will generate a good ranking of items. The item side information is integrated into the model through a trainable matrix, which is crucial when the data is highly sparse. The main advantage of this approach is that the model can be generalized to any recommendation scenario where a graph neural network is used for collaborative filtering. Experimental results obtained using the recent RS datasets show that the proposed model outperformed the related baselines. © 2024 Elsevier Ltd

Traditional recommender systems find similarities among users and items to generate recommendations. These systems have various advantages, such as no training process is involved, and they can generate recommendations by simply analyzing the users' past behaviors. These systems typically list suggestions based on similarity information of users and/or items. However, relying on only user-user or item-item similarity leads to a poor recommendation process, specifically in the case of sparse datasets. Further, the method used for obtaining similarity plays a crucial role in the performance of the recommender systems. Hence, in this study, we have integrated the different similarity measures, such as Pearson correlation and cosine similarity, to generate even better recommendations than the traditional user-user or item-item-based recommendation process. Experimental results obtained using the proposed approach for the benchmark dataset show significant improvement in the recommendation quality.

-

IoT devices depend on a network for communication and convenient information sharing. However, networks suffer from security threats such as DoS attack, RPL attack. We propose a honeypot-based method to detect such attacks. In proposed solution, Cowrie honeypot is used to capture adversaries activities. Captured unlabeled data is analyses by a modified Naive Bayes algorithm in combination with K-means clustering approach. Proposed solution effectively identifies DoS attacks and helps to strengthen the firewall and intrusion detection system functionalities.

In designing modern recommender systems, item feature information (or side information) is often ignored as most models focus on exploiting rating information. However, the side information is equally essential for capturing users' interests in items. Also, the recommender systems that use side information partially process the feature information by ignoring the locality-preserving property of item features. This study proposes an approach for collaborative filtering by applying an ordinal consistency-based matrix factorization (MF) model to maintain the locality-preserving property of item features to counter this problem. The ordinal consistency condition is implied using a loss function to the item features. Using MF removes the redundancy and inconsistency in item features, producing good results in calculating similarity information for recommendations. We have used five benchmark datasets to evaluate and compare the proposed model. Results obtained using the experiments suggest significant improvement in performance compared to related baselines.

Recommender Systems are helpful to many by filtering the information according to an individual’s preferences. However, the choice of a person may change with time. Keeping track of these changes or predicting the next sequence of items is difficult for any recommender system. Most time-aware recommender systems exploit users’ rating distribution to predict the next possible set of items. But due to the sparsity (users’ preferences) in real-world datasets, these models suffer from low accuracy. To overcome such issues, this study proposes a collaborative filtering model which uses a preference estimation technique that uses user and item feature information along with the ratings. Further, we have used preference relations (instead of ratings) to improve the ranking quality of the recommender system. We have used the ML-1M, NetFlix, and Last.fm datasets for evaluation, the benchmark datasets for recommender systems testing. All these three datasets have sparsity levels above 92%. Experimental results indicate significant improvement (5 to 7%) over the state-of-the-art related baselines.

A content-based recommender system uses essential item features that play a crucial role in building quality user preference profiles. However, in most real-world datasets, the item features are highly inconsistent and sparse, making it challenging to develop efficient user profiles. Additionally, the user preference profiles created by individual learners fail to learn from the misclassification of user ratings and preferences. Thus, to resolve these problems, this paper suggests a two-fold approach to improve the performance of the content-based recommender systems. The first approach is the refinement of the existing sparsity and inconsistencies in item features using matrix factorization. The second approach is the generation of individual preference profiles using iterative boosting of multiple weak learners for penalizing the misclassification of ratings. The suggested method is tested via benchmark recommender system datasets such as ML-1M, Last.fm, and Netflix. The results obtained during experiments show a significant improvement in recommendation quality over the state-of-the-art content-based recommender system models.