The paper introduces DEEPWALK, an algorithm that learns social representations of vertices in a graph using deep learning techniques. These representations capture neighborhood similarity and community membership, encoding social interactions in a continuous vector space. The method is applied to the Karate graph, demonstrating its ability to preserve community structure. DEEPWALK is evaluated on multi-label classification tasks on large heterogeneous graphs, showing improved performance over other methods, especially in sparse datasets. The algorithm is scalable and can be parallelized, making it suitable for web-scale graphs. The paper also discusses related work in relational learning and unsupervised feature learning, highlighting the unique contributions of DEEPWALK.The paper introduces DEEPWALK, an algorithm that learns social representations of vertices in a graph using deep learning techniques. These representations capture neighborhood similarity and community membership, encoding social interactions in a continuous vector space. The method is applied to the Karate graph, demonstrating its ability to preserve community structure. DEEPWALK is evaluated on multi-label classification tasks on large heterogeneous graphs, showing improved performance over other methods, especially in sparse datasets. The algorithm is scalable and can be parallelized, making it suitable for web-scale graphs. The paper also discusses related work in relational learning and unsupervised feature learning, highlighting the unique contributions of DEEPWALK.