Abstract: System and methods are provided that can address cold-start problems in database keyword searches. The search system generates machine-learned values for new item and queries based on historical signals for already existing item and queries. The values are used as input in a ranking model to rank search results for a user query. The initial values for the new item query pairs predict user engagement with the new item query pairs based on historical data for existing item query pairs and increase the visibility of new items to accumulate user interaction data for the new items. After additional user interactions are received, the values are updated using a Bayesian formula.

Abstract: Various embodiments provide for selecting a subset of features to use to train a model for search applications. To select a feature, the candidate features are randomly assigned into two groups. Each of the two groups represents a summation of the respective features that were assigned to it. Then a decision tree building scan is performed on the two groups to determine which of the two groups performs better based a selection criteria. Upon determining which of the two groups is better, the candidate features of the winning group are again randomly assigned into two groups. These two groups are again scanned as described above to determine a winning group. This binary splitting and scanning pattern is continuously performed until the winning group contains one remaining feature. That remaining feature is then designated as a selected feature to be used in the search model.