Optimized graph traversal

A system comprising:

• a property selector of a front end server that is configured to:
• receive training data that includes previous phrases submitted by digital component providers and content selected in response to the submitted phrases;
• a knowledge graph traverser that is configured to:
• determine, by accessing a knowledge graph, an entity that is being referred to by each of the previous phrases submitted by the digital component providers, wherein the knowledge graph comprises nodes that are each uniquely identify different entities and that are connected by relationships between connected nodes;
• and
• determine, by accessing the knowledge graph and for each entity referred to by each of the previous phrases, related entities that are within a threshold distance of a node corresponding to the entity,
• wherein the property selector is configured to:
• train, using machine learning, a selection model using the submitted phrases, the content selected in response to the submitted phrases, the entities referred to by each previous phrase, the related entities, and the relationship between the entities referred to by each previous phrase and the related entities, and
• wherein the front end server is configured to:
• receive, through a user interface, a phrase submitted by a digital component provider;
• provide the phrase as an input to the selection model;
• receive, from the selection model, a group of phrases that are related to the phrase;
• and
• update the user interface to present at least a portion of the group of phrases with programmatic controls that assign one or more of the phrases of the group of phrases as distribution criteria for digital components of the digital component provider in response to activation of the programmatic controls.

A computer-implemented method comprising:

• receiving training data that includes previous phrases submitted by digital component providers and content selected in response to the submitted phrases;
• determining, by accessing a knowledge graph, an entity that is being referred to by each of the previous phrases submitted by the digital component providers, wherein the knowledge graph comprises nodes that are each uniquely identify different entities and that are connected by relationships between connected nodes;
• determining, by accessing the knowledge graph and for each entity referred to by each of the previous phrases, related entities that are within a threshold distance of a node corresponding to the entity;
• training, using machine learning, a selection model using the submitted phrases, the content selected in response to the submitted phrases, the entities referred to by each previous phrase, the related entities, and the relationship between the entities referred to by each previous phrase and the related entities;
• receiving, through a user interface, a phrase submitted by a digital component provider;
• providing the phrase as an input to the selection model;
• receiving, from the selection model, a group of phrases that are related to the phrase;
• and
• updating the user interface to present at least a portion of the group of phrases with programmatic controls that assign one or more of the phrases of the group of phrases as distribution criteria for digital components of the digital component provider in response to activation of the programmatic controls.

A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

• receiving training data that includes previous phrases submitted by digital component providers and content selected in response to the submitted phrases;
• determining, by accessing a knowledge graph, an entity that is being referred to by each of the previous phrases submitted by the digital component providers, wherein the knowledge graph comprises nodes that are each uniquely identify different entities and that are connected by relationships between connected nodes;
• determining, by accessing the knowledge graph and for each entity referred to by each of the previous phrases, related entities that are within a threshold distance of a node corresponding to the entity;
• training, using machine learning, a selection model using the submitted phrases, the content selected in response to the submitted phrases, the entities referred to by each previous phrase, the related entities, and the relationship between the entities referred to by each previous phrase and the related entities;
• receiving, through a user interface, a phrase submitted by a digital component provider;
• providing the phrase as an input to the selection model;
• receiving, from the selection model, a group of phrases that are related to the phrase;
• and
• updating the user interface to present at least a portion of the group of phrases with programmatic controls that assign one or more of the phrases of the group of phrases as distribution criteria for digital components of the digital component provider in response to activation of the programmatic controls.