METHOD AND SERVER FOR DETERMINING TRAINING SET FOR MACHINE LEARNING ALGORITHM (MLA) TRAINING Russian patent published in 2024 - IPC G06N20/00 G06F16/906 

FIELD: physics.

SUBSTANCE: invention relates to a method and a server for determining a training set for training a machine learning algorithm (MLA) for classifying digital objects. Method comprises obtaining by server of multiple training examples for training MLA, training example includes text data associated with corresponding digital object, and indication of true class of corresponding object; server ordering said plurality of training examples into an ordered sequence of training examples, said training example has previous training examples in an ordered sequence and subsequent training examples in an ordered sequence; generating, by the server, a text feature for said training example based on the text data in said training example, as well as text data and true classes of only previous training examples in an ordered sequence without taking into account text data in subsequent training examples; determining, by a server, a training set for MLA based on said training example, wherein the training set has training input and a label, the training input includes a text feature, the label represents the true class of the corresponding object, wherein the digital object is any of: a digital document provided as a search result in response to a search query, digital element recommended to the user of the content recommendation system, e-mail message intended for an e-mail platform user.

EFFECT: high reliability of the prediction model by reducing the risk and effect of retraining during the phase of using the prediction model.

56 cl, 8 dwg