Abstract:The rapid growth of online social networks has underscored the importance of understanding the intensity of user relationships, referred to as "tie strength." Over the past few decades, extensive efforts have been made to assess tie strength in networks. However, the lack of ground-truth tie strength labels and the differing perspectives on tie strength among researchers have complicated the development of effective prediction methods for real-world applications. In our study, we first categorize mainstream understandings of tie strength into seven standardized definitions and verify their effectiveness by investigating the class distributions and correlations across these definitions. We also draw key insights into tie resilience from the perspective of tie dissolution that (1) stronger ties are more resilient than weaker ones, and (2) this tie resiliency ratio increases as the network evolves. We then conduct extensive experiments to evaluate existing tie strength prediction methods under these definitions, revealing that (1) neural network methods capable of learning from semantic features hold great potential for high performance, (2) models struggle under definitions that offer limited understandings of tie strength in the network, (3) existing models face imbalance issues that cannot be addressed by traditional quantity imbalance techniques, and (4) different definitions of tie strength allow for the inference of not only the current state but also the future state of a tie. Building on these findings, we propose strategies to improve existing methods and suggest several promising directions for future research. Code and datasets are provided at <a class="link-external link-https" href="https://github.com/XueqiC/tie_strength_prediction" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to understand and predict the strength of user relationships in social networks (i.e., "relationship strength" or "tie strength"). Specifically, the paper aims to: 1. **Standardize the Definition of Relationship Strength**: Since different researchers have different understandings and definitions of relationship strength and lack real - label data, it is difficult to develop effective prediction methods. Therefore, the paper first summarizes the mainstream definitions of relationship strength into seven standardized definitions and verifies the effectiveness of these definitions. 2. **Evaluate Existing Prediction Methods**: Based on the above - mentioned standardized definitions, the paper conducts extensive experimental evaluations of existing relationship - strength prediction methods, revealing the performance characteristics and limitations of existing models under different definitions. For example: - Neural network methods perform well when they can learn from semantic features. - The performance of models under certain definitions is limited by insufficient understanding of relationship strength. - Existing models face the imbalance problem, and traditional methods cannot solve it effectively. 3. **Propose Improvement Suggestions and Future Directions**: According to the experimental results, the paper proposes strategies to improve existing methods and points out several promising research directions to improve the accuracy and applicability of relationship - strength prediction. ### Core Contributions of the Paper - **Standardized Definitions**: Summarize and propose 7 standardized definitions of relationship strength as pseudo - real labels. - **Analysis and Verification**: Verify the effectiveness of these definitions by analyzing class distribution and correlation, and explore the resilience of strong and weak relationships from the perspective of relationship dissolution. - **Experimental Evaluation**: Conduct extensive experiments on existing prediction methods under multiple settings and develop new perspectives of understanding. - **Improvement Suggestions**: Based on the experimental results, propose strategies to improve existing models and point out future research directions. ### Main Findings - **Strong Relationships Are More Resilient**: Strong relationships are less likely to break than weak relationships, and as the network develops, the proportion of this resilience will increase. - **Great Potential of Neural Networks**: Neural network models that can utilize semantic features show significant advantages under certain definitions. - **Definitions Affect Performance**: Different definitions of relationship strength have a significant impact on model performance, and it is difficult for models under certain definitions to obtain good prediction results. - **Data Imbalance Problem**: Existing models face challenges when dealing with imbalanced data, and the effects of traditional imbalance - handling techniques are limited. Through these studies, the paper provides new insights into understanding and predicting the relationship strength in social networks and lays the foundation for future research.

A Comprehensive Analysis of Social Tie Strength: Definitions, Prediction Methods, and Future Directions

TranGAN: Generative Adversarial Network Based Transfer Learning for Social Tie Prediction.

Predicting tie strength with social media

The Social Bow Tie

Measuring Strength of Ties in Social Network

TieVis: visual analytics of evolution of interpersonal ties

Tie Strength in Online Social Networks and its Applications: A Brief Study

NEW: A generic learning model for tie strength prediction in networks

A Machine Learning Approach to Predicting Continuous Tie Strengths

Dimension Exploitation and Measurement of Tie Strength in Enterprise Network

Going beyond communication intensity for estimating tie strengths in social networks

Estimating tie strength in social networks using temporal communication data

Predicting Tie Strength with Ego Network Structures

Predicting Tie Strength of Chinese Guanxi by Using Big Data of Social Networks

Information Propagation in Online Social Networks: a Tie-Strength Perspective

Capturing Tie Strength with Algebraic Topology

Investigating and Modeling the Dynamics of Long Ties

From acquaintance to best friend forever: robust and fine-grained inference of social tie strengths

Social Recommendation with Strong and Weak Ties

A Longitudinal Social Network Clustering Method Based on Tie Strength.

Social Recommendation Using Quantified Social Tie Strength.