System and method for content discovery

A method of processing a data query for semantically similar broadcast channels associated with respective linear schedules of media content based on a defined query attribute type and value, the data query identifying the channel type as a target attribute type, the method comprising:

• a) for each of a plurality of broadcast channels:
• retrieving metadata of media content within a scheduling window of the broadcast channel,
• generating a feature vector from the retrieved metadata of the defined attribute type,
• selecting first and second sub-models from a trained artificial neural network model comprising an input layer consisting sets of input nodes corresponding to respective attribute types in media content metadata, each input node in a set corresponding to a respective attribute value of one of a plurality of predefined attribute types, and a hidden layer consisting each connected to a plurality of hidden nodes, each hidden node connected to every input node of the input layer and having a corresponding weight embodying a strength of connection therebetween, wherein the trained weights of each set of input nodes define a respective sub-model of the neural network that is associated with a corresponding attribute type in the media content metadata, each sub-model consisting a hidden state matrix of trained weights between the corresponding set of input nodes and said plurality of hidden nodes;
• combining the hidden state matrices of the selected first and second sub-models to determine a combined hidden state matrix consisting a plurality of hidden state vectors associated with a corresponding query attribute value;
• determining the hidden state vector from the combined hidden state matrix corresponding to the identified query attribute value from the query data;
• computing the cosine similarity between the feature vector and the determined hidden state vector to produce a score indicating relevancy of the broadcast channel to the query attribute value;
• b) sorting the plurality of computed scores indicating relevancy of the plurality of broadcast channels to the query attribute value;
• and
• c) outputting data identifying a ranked listing of broadcast channels based on the sorting plurality of computed scores.

An apparatus comprising one or more computer processors configured to process a request for content assets, the request identifying an attribute value of a query attribute type, by:

• retrieving, from a memory, first and second sub-models of a trained neural network, the first sub-model corresponding to a query attribute type and the second sub-model corresponding to a target attribute type, wherein data defining the trained neural network includes an input layer consisting sets of input nodes corresponding to respective attribute types in media content metadata, each input node in a set corresponding to a respective attribute value of one of a plurality of predefined attribute types, and a hidden layer consisting each connected to a plurality of hidden nodes, each hidden node connected to every input node of the input layer and having a corresponding weight embodying a strength of connection therebetween, wherein the trained weights of each set of input nodes define a respective sub-model of the neural network that is associated with a corresponding attribute type in the media content metadata, each sub-model consisting a hidden state matrix of trained weights between the corresponding set of input nodes and said plurality of hidden nodes;
• determining one or more relevancy parameters for the user by:
• combining the hidden state matrices of the retrieved first and second sub-models to determine a combined hidden state matrix consisting a plurality of hidden state vectors associated with a corresponding query or target attribute value;
• determining the hidden state vector from the combined hidden state matrix corresponding to the attribute value identified in the request, the determined hidden state vector defining parameters embodying predicted relevancy between the query attribute type and the target attribute type;
• determining the hidden state vector from the combined hidden state matrix corresponding to the attribute value identified in the request, the hidden state vector defining parameters embodying predicted relevancy between the query attribute type and the target attribute type for the user;
• and determining one or more relevancy parameters for the user based on the determined hidden state vector;
• and
• selecting one or more content assets for presentation by a media receiver device, wherein the or each selected content asset is based on a determined relevancy parameter.

A non-transitory computer-readable medium comprising computer-executable instructions that when executed perform the method of processing a request for content assets, the request identifying an attribute value of a query attribute type, by:

• retrieving, from the memory, first and second sub-models of the trained neural network, the first sub-model corresponding to a query attribute type and the second sub-model corresponding to a target attribute type, wherein data defining the trained neural network includes an input layer consisting sets of input nodes corresponding to respective attribute types in media content metadata, each input node in a set corresponding to a respective attribute value of one of a plurality of predefined attribute types, and a hidden layer consisting each connected to a plurality of hidden nodes, each hidden node connected to every input node of the input layer and having a corresponding weight embodying a strength of connection therebetween, wherein the trained weights of each set of input nodes define a respective sub-model of the neural network that is associated with a corresponding attribute type in the media content metadata, each sub-model consisting a hidden state matrix of trained weights between the corresponding set of input nodes and said plurality of hidden nodes;
• determining one or more relevancy parameters for the user by:
• combining the hidden state matrices of the retrieved first and second sub-models to determine a combined hidden state matrix consisting a plurality of hidden state vectors associated with a corresponding query or target attribute value;
• determining the hidden state vector from the combined hidden state matrix corresponding to the attribute value identified in the request, the determined hidden state vector defining parameters embodying predicted relevancy between the query attribute type and the target attribute type;
• determining the hidden state vector from the combined hidden state matrix corresponding to the attribute value identified in the request, the hidden state vector defining parameters embodying predicted relevancy between the query attribute type and the target attribute type for the user;
• and determining one or more relevancy parameters for the user based on the determined hidden state vector;
• and
• selecting one or more content assets for presentation by a media receiver device, wherein the or each selected content asset is based on a determined relevancy parameter.