[Paper Review] Context-Aware Service Recommendation System for the Social Internet of Things

Amar Khelloufi, Huansheng Ning+3 more|arXiv (Cornell University)|Aug 14, 2023

Digital Marketing and Social MediaSocial Sciences3 citations

TL;DR

This paper proposes CASR-SIoT, a context-aware service recommendation system for the Social Internet of Things that integrates review-based and engagement-based feature learning using Factorization Machines to model higher-order feature interactions. The framework improves recommendation accuracy by 12.5% in recall and 19.46% in precision over baseline methods by effectively capturing contextual and latent feature interactions in device-service pairs.

ABSTRACT

The Social Internet of Things (SIoT) enables interconnected smart devices to share data and services, opening up opportunities for personalized service recommendations. However, existing research often overlooks crucial aspects that can enhance the accuracy and relevance of recommendations in the SIoT context. Specifically, existing techniques tend to consider the extraction of social relationships between devices and neglect the contextual presentation of service reviews. This study aims to address these gaps by exploring the contextual representation of each device-service pair. Firstly, we propose a latent features combination technique that can capture latent feature interactions, by aggregating the device-device relationships within the SIoT. Then, we leverage Factorization Machines to model higher-order feature interactions specific to each SIoT device-service pair to accomplish accurate rating prediction. Finally, we propose a service recommendation framework for SIoT based on review aggregation and feature learning processes. The experimental evaluation demonstrates the framework's effectiveness in improving service recommendation accuracy and relevance.

Motivation & Objective

• To address the gap in existing SIoT recommendation systems that overlook contextual presentation of service reviews and fail to model latent feature interactions.
• To enhance recommendation accuracy by combining review-based and engagement-based feature learning for device-service pairs.
• To evaluate the effectiveness of a selective layer in extracting relevant semantic information from user reviews and service documents.
• To investigate the impact of hyperparameter settings on recommendation performance in dynamic SIoT environments.
• To develop a framework that integrates social relationships, contextual data, and multi-modal features for personalized service recommendations.

Proposed method

• Proposes a latent features combination technique to aggregate device-device relationships and capture latent feature interactions within the SIoT.
• Employs Factorization Machines to model higher-order feature interactions specific to each device-service pair for accurate rating prediction.
• Introduces a review-based feature learning component with a selective layer to extract and prioritize relevant semantic information from user reviews and service documents.
• Combines engagement-based and review-based feature learning to create comprehensive device-service representations.
• Utilizes a knowledge graph-like structure to model social relationships and preferences among devices and users in the SIoT.
• Employs a hybrid learning architecture that fuses multi-modal data, social correlations, and contextual factors for improved recommendation relevance.

Experimental results

Research questions

• RQ1 How does the integration of review-based and engagement-based feature learning affect the performance of service recommendation in SIoT?
• RQ2 What is the impact of different hyperparameter settings on the accuracy and effectiveness of the recommendation system?
• RQ3 How effective is the selective layer in identifying relevant semantic information from device-service pairs in the review-based feature learning process?
• RQ4 To what extent does the proposed framework outperform existing baseline models in terms of precision, recall, and F1-score across diverse datasets?
• RQ5 How well does the framework handle dynamic and sparse data in large-scale SIoT environments?

Key findings

• The CASR-SIoT framework outperformed baseline methods across all three datasets, achieving a 12.50% improvement in recall and a 19.46% improvement in precision compared to the LR-based baseline.
• The removal of the selective layer in the review-based feature learning component led to significant performance degradation, confirming its importance in identifying relevant semantic features.
• The FM-based model achieved the lowest performance (F1: 0.7423), indicating that Factorization Machines alone are insufficient without integrated feature learning components.
• The framework demonstrated superior performance in the Appliances, Cellphones, and Electronics categories, with F1-scores of 0.874, 0.812, and 0.831 respectively under the proposed method.
• The integration of both engagement-based and review-based feature learning significantly enhanced the system’s ability to model context-aware interactions and improve recommendation relevance.
• The results confirm that context-aware modeling of service reviews and social relationships leads to more accurate and personalized recommendations in dynamic SIoT environments.