The following diagram presents a high-level overview of the data model implemented in the Veritone GraphQL API.
Typically, each block in the diagram has a corresponding type in the GraphQL schema. Each type has corresponding queries that retrieve it and mutations to modify it.
Taking the Task and Job blocks from the diagram, we have the following
schema in GraphQL:
type Job {
# Retrieve all tasks contained in this job
tasks: [Task]
}
type Task {
# Retrieve the job that contains this task
job: Job
}
type Query {
# Retrieve a single job by ID
job(id: ID!): Job
# Retrieve a list of jobs
jobs: [Job]
# Retrieve a single task by ID
task(id: ID!): Task
}Note that it's been simplified, with only basic fields and no filtering or paging.
The following section describes what each entity means.
- Asset: a piece of data. An asset can be consumed by an engine for analysis, or it can be produced by an engine to store results.
- Temporal Data Object: contains assets and associated time series and other metadata.
- Engine: a self-contained, encapsulated program that is integrated into the Veritone platform and performs either AI analysis or infrastructure functions such as ingestion. Third-party developers in the Veritone ecosystem can create, deploy, and market their own engines using the Veritone Developer Application (VDA).
- Task: a task is a request to run a single engine. A task is created with a payload containing input to the engine.
- Job: a job describes all the tasks necessary to perform a single high-level operation, such as run a series of engines against a piece of data or ingest data from a given source.
- Organization: a Veritone platform subscriber. An organization may represent a company, a department within a company, or an individual developer.
- User: a single user account within an organization, representing a human user (not a service or application).
- Role: an abstract, business-focused way of describing a user's permissions on the Veritone platform. For example, a "Library Editor" can manage libraries on behalf of the organization. A role collects a set of fine-grained permissions.
- Permission: a single functional permission on the Veritone platform, such as "create library" or "view asset".
- Application: a custom Veritone client application. Third-party developers in the Veritone ecosystem can create, deploy, and market their own applications using the Veritone Developer Application (VDA).
- Engine Category: a grouping of engines with similar features. Expected engine payload and output are common to engines within an engine category. Engine categories make it easier for users to find and select the engines they want to use.
- Engine Whitelist / Engine Blacklist: users within a given organization are allowed to search, view, and use a set of engines determined by the engine whitelist and engine blacklist for the organization. These lists are managed by Veritone in accordance with the organization's needs.
- Build: engine developers upload an engine build to deploy or update their engine; a build represents the engine code. Each engine can have any number of builds representing different versions of the code, only one of which is active at any given time.
- Library: A named collection of entities an organization is interested in identifying in media, e.g. American Politicians. A library's type defines what type of entities it can hold.
- Library type: Describes the type of entities the library contains. Used to tailor the UI experience for specific types of entities (for example, People, Ads, etc).
- Library collaborator: An external organization that a library has been shared with. Users in the collaborator organization can view the library and use it to train their own models, but cannot modify it.
- Library engine model: Data generated during an trainable engine's training step. The model is (in some cases) provided to an engine when the engine is run. A model can optionally contain an asset representing the model data.
- Entity: An aggregation of assets (entity identifiers) for a defined concept, such as a person, a company or organization, an advertising campaign, a type of object, etc.
- Entity identifier: An asset associated with an entity. Examples are headshots for face recognition, voice clips for speaker recognition, ad creative for audio fingerprinting, DLM for transcription, and aliases for transcription.
- Entity identifier type: The type of asset associated with an entity. e.g. headshot, logo, voice clip, ad, DLM. The library type defines what identifier types it can support.
- Mention: A record that a given entity, tag, or search condition was matched (or "mentioned") in an engine result. Mentions can be shared and published.
- Mention comment: A user comment on an individual mention. Comments are used to share, publicize, and collaborate on mentions.
- Mention rating: A user rating on an individual mention. Ratings are used to share, publicize, and collaborate on mentions.
- Watchlist: An enhanced stored search including a set of search conditions and filters, effective start and stop date, and other information. Hits against the search captured in a watchlist generate mentions.
- Folder: A folder is a container for organizing and sharing information, including temporal data objects (media and other data), watchlists, other folders.
- Root folder: an organization has an implicit top-level folder for each folder type (collection, watchlist, etc.).
- Collection: a group of related data. A collection can be placed in a folder, or not. Collections can be shared and publicized.
