Addressing data scarcity in early-cycle annual software services

Abstract

Certain aspects of the disclosure provide a method for training a machine-learning model for predicting content recommendations. The method may include obtaining multi-year training data associated with the software subscription service, the multi-year training data comprising features and content relevance labels for pre-authentication activities and post-authentication activities; filtering the multi-year training data based on a timestamp associated with a respective content relevance label of the content relevance labels such that each respective content relevance label is associated with features having timestamps occurring at or before a timestamp of the respective content relevance label; and training a machine-learning model using the filtered multi-year training data to predict content recommendations based on an aggregated multi-year feature set, wherein the aggregated multi-year feature set includes features extracted from current year pre-authentication activities and post-authentication activities aggregated with historical features extracted from pre-authentication activities and post-authentication activities from prior years.