Research On Graph Neural Network Recommendation Algorithm With Joint Feature Interaction

With the rapid development of information technology,a vast amount of interaction data has been generated between various applications/services and users.The challenge for researchers and technology developers lies in excavating potential user preferences and customizing personalized recommendations from this massive data.Existing recommendation algorithms analyze users’ historical records to recommend items of interest.However,two problems exist : the underutilization of rich item feature data and the inability to directly reveal the association between users and item features.To address these issues,this paper proposes a novel recommendation algorithm called Attention Interaction Graph Collaborative Filtering(ATGCF)that incorporates joint feature interactions.The main contributions of this work are as follows:1、Introducing the ATGCF model,which captures users’ potential preferences for multi-feature items by modeling the relationships between users and item features,building on attention-based feature interactions in graph collaborative filtering.The ATGCF model consists of two parts:a user-item feature interaction layer that utilizes a feature interaction network constructed with multi-head attention and fully connected layers to capture users’ preferences for multiple features,moreover,a graph neural collaborative filtering layer that analyzes user-item interaction history to construct a heterogeneous bipartite graph of user-feature-item and designs an aggregation neural network to capture the high-order connectivity among users,items,and item features.To validate the effectiveness and advancedness of the ATGCF model,experiments are conducted on three public datasets,and the results show that the ATGCF algorithm outperforms other baseline indicators by 4.43% in recall,5.56% in precision,and 5.16%in normalized discounted cumulative gain(NDCG)evaluation.Furthermore,additional experiments demonstrate that the ATGCF model also achieves good recommendation performance when it works on small-sample datasets.2、To validate the effectiveness of the feature graph recommendation model,a joint graph recommendation method is proposed for the computation offloading prediction in the current offloading scenario.Based on the similarities between task features and edge device features,task nodes and edge device nodes are abstracted as user nodes and item nodes,respectively,in the recommendation task.Then,using the historical offloading records of tasks with similar features,the target offloading node for the current task is predicted.Experimental analysis is conducted to validate the effectiveness and advancedness of the proposed method.The results demonstrate that the proposed joint graph recommendation method effectively addresses the challenges of computation offloading.