Every graph can be represented as an adjacency matrix. An adjacency matrix is a square matrix where the elements indicate whether pairs of nodes are connected. Such a matrix can be regarded as a high-dimensional representation of the network. Now imagine you have a graph with a million nodes. In an adjacency matrix, each node would be represented with a row in the matrix that has a million elements. Suppose you want to train a machine learning model and somehow use the network structure information as an input feature. Using the adjacency matrix as an input to your model could cause all sorts of problems, from having too many input features to overfitting. In practice, you often want to embed a node's local representation to compare nodes with similar neighborhood topology instead of using all relationships between nodes as a feature input. Node embedding techniques try to solve these issues by learning lower-dimensional node representation for any given network. The learned node representations or embeddings should automatically encode the network structure so that the similarity in the embedding space approximates the similarity in the network. Additionally, knowledge graph embedding models aim to encode both the nodes as well as the relationships in the embedding space.
Speakers: Tomaz Bratanic
Format: Full Session 30-45 min
Level: Advanced
Topics: #GraphDataScience, #Analytics, #General, #Advanced
Region: EMEA
Slides: https://dist.neo4j.com/nodes-20202-slides/096%20From%20Node%20to%20Knowledge%20Graph%20Embeddings%20-%20NODES2022%20EMEA%20Advanced%207%20-%20Tomaz%20Bratanic.pptx
Visit https://neo4j.com/nodes-2022 learn more at https://neo4j.com/developer/get-started and engage at https://community.neo4j.com