اسما رشيديان

Community analysis in social networks is a significant an‎d practical topic across various domains in todayʹs world. Link prediction, with applications such as targeted advertising, identification of influential individuals, disease propagation modeling, community detection, an‎d other purposes, is a well-established research area. Given the influence of social network communities on individualsʹ choices an‎d behaviors, predicting usersʹ joining of communities in these networks is of paramount importance. The objective of this thesis is to propose a method for predicting usersʹ joining of communities in social networks. Social networks are modeled as graphs for analytical purposes. While numerous studies have addressed link prediction, focusing on the connection between one node an‎d another, an‎d extensive research has been conducted on community analysis an‎d predicting their evolution, no study has specifically targeted the prediction of usersʹ joining of communities. Therefore, in this thesis, the problem is first formally defined, an‎d a method named UCJP (User-to-Community Join Prediction) is proposed to predict usersʹ joining of communities. In the UCJP method, communities are first detected, an‎d then structural features of the communities, along with the spatial embedding features of nodes in the network that change over time, are extracted to form a feature vector. Subsequently, the feature vector, combined with the vector of outlier data, is fed into recurrent an‎d graph neural networks, resulting in the construction of a predicted graph. The proposed method is eva‎luated by comparing the predicted graph with the actual graph over the specified prediction time frame. To eva‎luate the UCJP method, in addition to its implementation, five related methods (including four link prediction methods an‎d one community recommendation method adapted to the problem) were also executed. The implementation results demonstrate that the UCJP method achieves improvements in performance for predicting usersʹ joining of communities, with precision ranging from 26% to 59%, recall from 18% to 69%, accuracy from 24% to 59%, an‎d F1-score from 24% to 63% compared to related works. Notably, in the traditional link prediction problem (node-to-node), the UCJP method also outperforms 13 existing methods in the relevant domain.

پيش‌بيني پيوند , تحليل شبكه‌هاي اجتماعي , پيش‌بيني پيوستن كاربر به جوامع , جوامع در شبكه‌هاي اجتماعي , شبكه‌هاي عصبي

Link Prediction , Social Network Analysis , User-Community Joining Prediction , Communities in Social Networks , Neural Networks

Asma Rashidian

Dr. Hassan Naderi