Entity Embeddings is an unsupervised, deep learning technique developed by Google Brain researchers for representing structure in unstructured data. It is an extension of the deep learning technique known as word embeddings. While word embeddings map words to a multi-dimensional vector, entity embeddings map entities, such as entities in a knowledge graph, to a multi-dimensional vector. This vector space can then be used to represent relationships between entities or the entire knowledge graph in a single vector space.
The technique of entity embeddings was first introduced in a 2013 paper by researcher Tomas Mikolov and colleagues at Google Brain. In the paper, the authors proposed a method of learning distributed representations of entities in a knowledge graph that can be used for reasoning and inference without the need for supervised training. The technique has since been widely adopted, with applications in tailored search, natural language processing, and a variety of other fields.
The core of the Entity Embeddings technique is a neural network. The network takes as input a set of entities and their relationships and outputs a distributed representation of the knowledge graph. The relationships are represented as weighted edges between the entities. These edges are then used to calculate a score for each entity using the following equation: score = (Weight*Entity_Vector 1) + (Weight*Entity_Vector 2) + … + (Weight*Entity_Vector N). This score is used to represent the similarity between the entities and the strength of their relationship.
The advantage of using entity embeddings is that it allows for a more accurate and richer representation of a knowledge graph than other methods, such as supervised machine learning. By learning robust data representations without the need for manual annotation, entity embeddings can be used to more accurately capture patterns and associations in complex data.
The technique has become increasingly popular in recent years, with many leading companies, such as Google, Microsoft, and Facebook, utilizing the technology for various applications. As the technology matures, more applications and use cases for entity embeddings will emerge and the technique will become even more widely used.
