System and method for capturing seasonality and newness in database searches

A system comprising:

• one or more processors;
• and
• one or more non-transitory computer-readable media storing computing instructions that, when executed on the one or more processors, cause the one or more processors to perform functions comprising:
• receiving a first access history for a record during a first time period, the first access history indicating a first number of times the record was accessed within the first time period;
• receiving one or more second access histories for the record for one or more additional time periods different than the first time period, each respective second access history of the one or more second access histories indicating one or more respective additional number of times the record was accessed within each respective additional time period of the one or more additional time periods;
• generating a feature vector for the record based on the first access history and the one or more second access histories of the record, wherein the feature vector comprises a concatenated list of features ending with a label for the feature vector;
• inputting the feature vector for the record into a supervised machine learning model, wherein the supervised machine learning model is (1) configured to determine a likelihood of the record being accessed by retrieving two or more training feature vectors associated with an item, and (2) trained on the two or more training feature vectors, wherein the two or more training feature vectors comprise:
• at least one training feature vector associated with the item represented by the record;
• and
• at least one training feature vector specifying a respective error cost, wherein the respective error cost comprises a cost of a false positive and a cost of a false negative;
• wherein the supervised machine learning model is modified based on the two or more training feature vectors and a cost-insensitive supervised machine learning algorithm;
• minimizing the respective error cost for each training feature vector by modifying a respective cost vector associated with each respective error cost as processed over repeated multiple iterations until at least a false negative rate is reduced by a defined percentage rate;
• outputting, via the supervised machine learning model, a probability distribution of the likelihood of the record being accessed;
• and
• facilitating storing the record in one of a first database cluster H or a second database cluster L based on the likelihood of the record being accessed.