Author: Cyrus Vahid, Da Zheng, George Karypis and Balaji Kamakoti: AWS AI
Knowledge Graphs (KGs) have emerged as an effective way to integrate disparate data sources and model underlying relationships for applications such as search. At Amazon, we use KGs to represent the hierarchical relationships among products; the relationships between creators and content on Amazon Music and Prime Video; and information for Alexa’s question-answering service. Information extracted from KGs in the form of embeddings is used to improve search, recommend products, and infer missing information. It is however hard to train KG embeddings from graphs with millions of nodes and billions of edges.
Amazon recently launched DGL-KE, a software package that simplifies this process with simple command-line scripts. With DGL-KE, users can generate embeddings for very large graphs 2–5x faster than competing techniques. DGL-KE provides users the flexibility to select models used to generate embeddings and optimize performance by configuring hardware, data sampling parameters, and the loss function. To use this package effectively, however, it is important to understand how embeddings work and the optimizations available to compute them. This two-part blog series is designed to provide this information and get you ready to start taking advantage of DGL-KE.
What is a graph
A graph is a structure used to represent things and their relations. It is made of two sets — the set of nodes (also called vertices) and the set of edges (also called arcs). Each edge itself connects a pair of nodes indicating that there is a relation between them. This relation can either be undirected, e.g., capturing symmetric relations between nodes, or directed, capturing asymmetric relations. For example, if a graph is used to model the friendship relations of people in a social network, then the edges will be undirected as they are used to indicate that two people are friends; however, if the graph is used to model how people follow each other on Twitter, the edges will be directed. Depending on the edges’ directionality, a graph can be _directed _or undirected.
Graphs can be either _homogeneous _or heterogeneous. In a homogeneous graph, all the nodes represent instances of the same type and all the edges represent relations of the same type. For instance, a social network is a graph consisting of people and their connections, all representing the same entity type. In contrast, in a heterogeneous graph, the nodes and edges can be of different types. For instance, the graph for encoding the information in a marketplace will have buyer, seller, and product nodes that are connected via wants-to-buy, has-bought, is-customer-of, and is-selling edges.
Finally, another class of graphs that is especially important for knowledge graphs are multigraphs. These are graphs that can have multiple (directed) edges between the same pair of nodes and can also contain loops. These multiple edges are typically of different types and as such most multigraphs are heterogeneous. Note that graphs that do not allow these multiple edges and self-loops are called simple graphs.
What is a knowledge graph
In the earlier marketplace graph example, the labels assigned to the different node types (buyer, seller, product) and the different relation types (wants-to-buy, has-bought, is-customer-of, is-selling) convey precise information (often called semantics)about what the nodes and relations represent for that particular domain. Once this graph is populated, it will encode the knowledge that we have about that marketplace as it relates to types of nodes and relations included. Such a graph is an example of a knowledge graph.
A knowledge graph (KG) is a directed heterogeneous multigraph whose node and relation types have domain-specific semantics. KGs allow us to encode the knowledge into a form that is human interpretable and amenable to automated analysis and inference. KGs are becoming a popular approach to represent diverse types of information in the form of different types of entities connected via different types of relations.
When working with KGs, we adopt a different terminology than the traditional vertices and edges used in graphs. The vertices of the knowledge graph are often called _entities _and the directed edges are often called _triplets _and are represented as a (h, r, t) tuple, where _h _is the head entity, _t _is the tail entity, and _r _is the relation associating the head with the tail entities. Note that the term relation here refers to the type of the relation (e.g., one of wants-to-buy, has-bought, is-customer-of, and is-selling).
Let’s now examine a KG with cast of a people and the world in which they live.
#deep-learning #embedding #neural-networks #graph #machine-learning #deep learning
