Matrix completion and recommendation provision with deep learning

A system comprising:

• a pre-processing module implemented at least partially in hardware of at least one computing device to encode an incomplete matrix effective to generate an incomplete encoded matrix in which known values of the incomplete matrix are encoded for input to an autoencoder regardless of whether attributes represented by the incomplete matrix are numerical, categorical, or a combination of numerical and categorical, the incomplete matrix encoded based on:
• a numerical mask indicative of the attributes of the incomplete encoded matrix that correspond to identified numerical attributes, the numerical mask including a number of cells that corresponds to a number of columns of the incomplete encoded matrix, and each cell of the numerical mask indicating whether an attribute of a corresponding column of the incomplete encoded matrix is a numerical attribute or not;
• and
• a categorical mask indicative of the attributes of the incomplete encoded matrix that correspond to identified categorical attributes, the categorical mask including the number of cells, and each cell of the categorical mask indicating whether the attribute of the corresponding column of the incomplete encoded matrix is a categorical attribute or not;
• a machine-learning model manager implemented at least partially in the hardware of the at least one computing device to train an autoencoder to:
• receive as input the incomplete encoded matrix;
• receive as additional input a missing value mask having a same number of cells as the incomplete encoded matrix, wherein each cell of the missing value mask indicates whether a corresponding cell of the incomplete encoded matrix has a known value or an unknown value;
• and
• output a complete encoded matrix by imputing encoded values in place of unknown values of the incomplete encoded matrix;
• and
• a post-processing module implemented at least partially in the hardware of the at least one computing device to decode the complete encoded matrix effective to generate a complete matrix in which imputed encoded values are converted to have a same format as the known values corresponding to a same attribute.

One or more non-transitory computer-readable storage media storing instructions that are executable by one or more processors to perform operations including:

• encoding an incomplete matrix to generate an incomplete encoded matrix in which known values of the incomplete matrix are encoded for input to an autoencoder regardless of whether attributes represented by the incomplete matrix are numerical, categorical, or a combination of numerical and categorical, the encoding including:
• generating a numerical mask indicative of the attributes of the incomplete encoded matrix that correspond to identified numerical attributes, the numerical mask including a number of cells that corresponds to a number of columns of the incomplete encoded matrix, and each cell of the numerical mask indicating whether an attribute of a corresponding column of the incomplete encoded matrix is a numerical attribute or not;
• and
• generating a categorical mask indicative of the attributes of the incomplete encoded matrix that correspond to identified categorical attributes, the categorical mask including the number of cells, and each cell of the categorical mask indicating whether the attribute of the corresponding column of the incomplete encoded matrix is a categorical attribute or not;
• providing the incomplete encoded matrix as input to the autoencoder;
• providing a missing value mask as additional input to the autoencoder, the missing value mask having a same number of rows and a same number of columns as the incomplete encoded matrix;
• and
• decoding a complete encoded matrix output by the autoencoder, the complete encoded matrix having encoded values imputed in real-time by the autoencoder in place of unknown values of the incomplete encoded matrix, and the decoding effective to generate a complete matrix in which imputed encoded values are converted to have a same format as the known values that correspond to a same attribute.