Hybrid Graph-Behavioral Predictive Framework For Trust-Aware And Secure Recommendation Systems

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The reliability and security of recommendation systems are now very important issues in the current digital era. ecosystems. Traditional collaborative filtering and hybrid methods fail to consider user. They are subject to data sparsity, shilling attacks, and behavioral noise, due to their trustworthiness. This study presents a Hybrid Graph-Behavial Predictive Framework, which combines Graph Convolutional To model social trust and behavioral relationship jointly within networks (GCN) using Temporal Attention Encoding. consistency. The framework is in two phases. In Stage 1, user trust levels are predicted to be based on Integrating graph-based relationships and temporal behavioural patterns for building dynamic trust scores. Trust, as a result of these trust scores, will be used in the recommendation process in stage 2, using a trust: Weighted attention mechanism, improving robustness by decreasing the impact of unreliable users. Experiments were run on the Epinions Trust Network data, which has 49,290 users and 487,181 trust relationships, and 664,824 rating interactions, with satisfactory results, and an accuracy of 95.65%, precision of 94.89%, recall of 93.18%, F1-score of 94.03%, and ROC-AUC of 0.9889 in trust prediction. The model obtains NDCG-0.93 and F1-score- 0.91 in a recommendation task. Furthermore, the attack detection rate is greatly enhanced from 45% to 92%, and overall The security of the system increases from 60% to 85%. The proposed framework provides a scalable, interpretable and A powerful approach to trust-aware recommendation system.