System and method for content discovery

Method of providing content using predictive control comprising; receiving S6-1 query for content items identifying attribute value of query attribute type; selecting S6-5, S6-9 first and second sub-models from trained artificial neural network model, corresponding to query and target attribute types respectively, each sub-model corresponding to an attribute type and defining trained weights between inputs nodes each representing an attribute value of the respective attribute type, and hidden nodes of the ANN; predicting user behaviour by combining S6-11 first and second sub-models; selecting S6-15 a subset of data from the combined sub-model using identified S6-13 attribute value from query as index to address the combined sub-model; sorting S6-17 the subset by a defined criterion, wherein the subset defines a ranked association between the query and target attribute types as parameter of predicted user behaviour; and outputting S6-15 data to control identification and provision of content items based on parameter. The sub-models may each be a two-dimensional matrix of trained weights, and combining sub-models may comprise multiplying the first matrix with a transpose of the second matrix. Used for content discovery in TV programme system. Also disclosed is a similar method of identifying relevant broadcast channels.