Link prediction in complex networks is a critical task in network analysis, aiming to identify missing or future links based on existing connections and node attributes. This survey summarizes recent advancements in link prediction algorithms, emphasizing contributions from physics perspectives, such as random-walk-based methods and maximum likelihood methods. It introduces three key applications: network reconstruction, evaluation of network evolution mechanisms, and classification of partially labeled networks. The article also outlines future challenges in link prediction. The link prediction problem involves determining the likelihood of a link between two nodes using observed links and node attributes. It is essential for analyzing networks with missing data and predicting future links in evolving networks. For example, in social networks, link prediction can help identify missing connections, while in biological networks, it can help detect spurious links. The article discusses various similarity-based algorithms, including local, global, and quasi-local indices. Local indices, such as Common Neighbors (CN), Salton Index, Jaccard Index, and Hub Promoted Index (HPI), measure the similarity between nodes based on their immediate neighbors. Global indices, such as Katz Index, Leicht-Holme-Newman Index (LHN2), and Average Commute Time (ACT), consider the entire network structure. Quasi-local indices, like Local Path Index (LP) and Local Random Walk (LRW), combine local and global information to improve prediction accuracy. Maximum likelihood methods, such as the hierarchical structure model and stochastic block model, are also discussed. These methods use probabilistic models to infer network structures and predict missing links. The hierarchical structure model infers hierarchical organization from network data, while the stochastic block model partitions nodes into groups and predicts link probabilities based on group memberships. Probabilistic models, including Probabilistic Relational Models (PRM), Probabilistic Entity Relationship Models (PERM), and Stochastic Relational Models (SRM), are introduced as tools for abstracting network structures and predicting missing links. These models optimize target functions to best fit observed data and estimate the probability of link existence. The article concludes by highlighting the importance of link prediction in various applications and the need for further research to improve prediction accuracy and efficiency. It emphasizes the role of physical approaches and probabilistic models in advancing the field of link prediction in complex networks.Link prediction in complex networks is a critical task in network analysis, aiming to identify missing or future links based on existing connections and node attributes. This survey summarizes recent advancements in link prediction algorithms, emphasizing contributions from physics perspectives, such as random-walk-based methods and maximum likelihood methods. It introduces three key applications: network reconstruction, evaluation of network evolution mechanisms, and classification of partially labeled networks. The article also outlines future challenges in link prediction. The link prediction problem involves determining the likelihood of a link between two nodes using observed links and node attributes. It is essential for analyzing networks with missing data and predicting future links in evolving networks. For example, in social networks, link prediction can help identify missing connections, while in biological networks, it can help detect spurious links. The article discusses various similarity-based algorithms, including local, global, and quasi-local indices. Local indices, such as Common Neighbors (CN), Salton Index, Jaccard Index, and Hub Promoted Index (HPI), measure the similarity between nodes based on their immediate neighbors. Global indices, such as Katz Index, Leicht-Holme-Newman Index (LHN2), and Average Commute Time (ACT), consider the entire network structure. Quasi-local indices, like Local Path Index (LP) and Local Random Walk (LRW), combine local and global information to improve prediction accuracy. Maximum likelihood methods, such as the hierarchical structure model and stochastic block model, are also discussed. These methods use probabilistic models to infer network structures and predict missing links. The hierarchical structure model infers hierarchical organization from network data, while the stochastic block model partitions nodes into groups and predicts link probabilities based on group memberships. Probabilistic models, including Probabilistic Relational Models (PRM), Probabilistic Entity Relationship Models (PERM), and Stochastic Relational Models (SRM), are introduced as tools for abstracting network structures and predicting missing links. These models optimize target functions to best fit observed data and estimate the probability of link existence. The article concludes by highlighting the importance of link prediction in various applications and the need for further research to improve prediction accuracy and efficiency. It emphasizes the role of physical approaches and probabilistic models in advancing the field of link prediction in complex networks.