Research On Knowledge Graph Guided Social Network Account Classification And Link Prediction Method

After the advent of the Information Age,human society has witnessed a continuous rise in the scale and prevalence of online social networks,which have become indispensable tools for personal communication and sharing.Due to the wealth of data contained within social networks,they provide a vast amount of real-time,high-value information and reflect patterns of human social activities,garnering significant attention from academia and industry alike.The classification of social network accounts and link prediction are crucial aspects of social network analysis and currently represent focal points of research in the field of social informatics.These two tasks contribute to the extraction of valuable information from social network data,facilitating account management,and providing substantial support for intelligent search,similarity recommendations,and social network administration.However,current research on social network account classification and link prediction suffers from issues such as limited consideration of information dimensions,insufficient completeness,and lack of interpretability.To address these challenges,this thesis proposes an approach that leverages and represents the rich heterogeneous information present in social networks.The main objectives and contributions of this work are summarized as follows:(1)To overcome the limitation of traditional social network analysis,which relies on a single type of graph structure data and fails to adequately represent the complex multidimensional information within social networks,this thesis investigates methods for constructing social network knowledge graphs.By extensively employing various analytical techniques and knowledge graph construction methodologies,we aim to establish comprehensive social network knowledge graphs that serve as a foundation for exploring rich associations between accounts.(2)In response to the inaccuracies resulting from incomplete and highly uncertain social network data,we propose a knowledge path-guided method for social network account classification.Introducing the concept of knowledge paths,we extract valuable path relationships within the social network and subsequently propose an account similarity metric based on these knowledge paths.Compared to traditional similarity metrics,our approach more accurately measures the degree of association between social accounts.Finally,to address the limitations of traditional graph convolutional networks in the context of social networks,we present a knowledge path-guided method for social network account classification,aiming to achieve more accurate account classification outcomes.(3)To overcome the issue of low utilization of heterogeneous association relationships within social networks,which results in suboptimal link prediction accuracy,we propose a knowledge path attention mechanism-based method for social network link prediction.By decomposing the social network knowledge graph into multiple account association views using various knowledge paths,we incorporate an attention mechanism to perform feature fusion across multiple views.Finally,we calculate link scores and make predictions.Our method leverages knowledge paths to extract multidimensional association relationships between accounts and automatically calculates the contribution of various relationships to the effectiveness of link prediction,thereby overcoming the expressive limitations of traditional heuristic algorithms.