Suggesting Resources using a Latency-Efficient Machine-Trained Ranking Model

A computing system for suggesting resources, comprising:

• an instruction data store for storing computer-readable instructions;
• a processing system for executing the computer-readable instructions in the data store, to perform operations comprising:
• receiving a query that is input via a search interface presentation;
• retrieving candidate resource descriptors, using a prefix tree data structure, that match the query, the candidate resource descriptors being data items that describe different kinds of candidate resources;
• generating feature information that describes the candidate resource descriptors, the feature information for a particular resource descriptor and an associated particular resource including:
• at least one feature that describes an extent to which the query matches the particular resource descriptor;
• and at least one feature that describes a resource type associated with the particular resource;
• using a machine-trained ranking model to generate scores associated with the candidate resource descriptors, based on the feature information, the scores enabling comparison across different types of resources;
• and
• generating output information that represents a set of resource descriptors selected based on the scores, for presentation via the search interface presentation.

A computer-readable storage medium for storing computer-readable instructions, a processing system executing the computer-readable instructions to perform operations, the operations comprising.

• receiving a query that is input via a search interface presentation;
• retrieving candidate resource descriptors, using a prefix tree data structure, that match the query, the candidate resource descriptors being data items that describe different kinds candidate resources;
• generating feature information that describes the candidate resource descriptors;
• using a machine-trained ranking model to generate scores associated with the candidate resource descriptors, based on the feature information;
• and
• generating output information that represents a set of resource descriptors that are selected based on the scores, for presentation via the search interface presentation,
• wherein the machine-trained ranking model includes an ensemble of plural component classification models, and wherein a training system produces machine-trained model by:
• producing plural candidate models based on a set of training examples, each having a particular number of component classification models and a particular tree depth;
• assessing a prediction time and an accuracy of each candidate model, to provide test results, the prediction time describing an amount of time that is required to map the feature information to an output result;
• and
• ranking the candidate models based on prediction times and accuracies associated with the candidate models, and selecting a candidate model that is ranked as most optimal.