Ranking search results using weighted topologies

A method comprising:

• receiving a plurality of topologies at a computing device, wherein each topology corresponds to a feature of a plurality of items;
• generating a weight for each topology at the computing device;
• receiving a search log at the computing device, wherein the search log comprises queries and identifiers of items selected from a results set presented in response to each query;
• computing a first distribution of the items selected in the search log by the computing device;
• computing a second distribution of the items using the topologies and the weights associated with each topology by the computing device;
• comparing the first and the second distributions by the computing device;
• adjusting one or more of the generated weights based on the comparison by the computing device;
• and
• providing the generated weights by the computing device.

A system comprising:

• at least one computing device;
• a search engine adapted to:
• receive a query;
• and
• generate identifiers of items in response to the query, wherein each item is associated with a plurality of features values and each feature value is associated with a feature of a plurality of features;
• and
• a ranker adapted to:
• receive the identifiers of items from the search engine;
• rank the identifiers of items using a plurality of topologies and weights, wherein each topology corresponds to a feature of the plurality of features, and each topology comprises transition probabilities between items for the feature values associated with the feature corresponding to the topology, and wherein the transition probability between a first item and a second item of a topology represents a probability that a preference for the first item will change to a preference for the second item based on the feature corresponding to the topology;
• and
• provide the ranked identifiers of items to the search engine.