Multi-task training process based on collaborative data
AssigneeMeta Platforms, Inc.
InventorsOren Sar Shalom, Yaakov Bibas
Life of this patent
Estimating the term…
Abstract
Systems, apparatuses and methods provide technology that generates an interaction matrix from collaboration data, where the collaboration data represents user interactions with a plurality of items. The technology applies a matrix factorization process to the interaction matrix to generate user vectors and matrix item vectors, and executes a multi-task training process with a neural network. The multi-task training process includes processing the collaboration data with the neural network to generate latent item vectors, and training the neural network based on a comparison of the latent item vectors to the matrix item vectors.
Claims17
At least one non-transitory computer readable storage medium comprising a set of instructions, which when executed by a computing device, cause the computing device to:
- generate an interaction matrix from collaboration data, wherein the collaboration data represents user interactions with a plurality of items;
- apply, prior to a multi-task training process with a neural network, a matrix factorization process to the interaction matrix to generate user vectors and matrix item vectors;
- and
- execute the multi-task training process with the neural network, wherein the multi-task training process includes:
- processing the collaboration data with the neural network to generate latent item vectors,
- identifying distances between the latent item vectors and the matrix item vectors;
- generating a first loss based on the distances;
- and
- training the neural network based on the first loss.
A system comprising:
- one or more processors;
- and
- a memory coupled to the one or more processors, the memory comprising instructions executable by the one or more processors, the one or more processors being operable when executing the instructions to:
- generate an interaction matrix from collaboration data, wherein the collaboration data represents user interactions with a plurality of items;
- apply, prior to a multi-task training process with a neural network, a matrix factorization process to the interaction matrix to generate user vectors and matrix item vectors;
- and
- execute a multi-task training process with the neural network, wherein the multi-task training process includes:
- processing the collaboration data with the neural network to generate latent item vectors,
- identifying distances between the latent item vectors and the matrix item vectors;
- generating a first loss based on the distances;
- and
- training the neural network based on the first loss.
A method comprising:
- generating an interaction matrix from collaboration data, wherein the collaboration data represents user interactions with a plurality of items;
- applying, prior to a multi-task training process with a neural network, a matrix factorization process to the interaction matrix to generate user vectors and matrix item vectors;
- and
- executing a multi-task training process with a neural network, wherein the multi-task training process includes:
- processing the collaboration data with the neural network to generate latent item vectors,
- identifying distances between the latent item vectors and the matrix item vectors;
- generating a first loss based on the distances;
- and
- training the neural network based on the first loss.
Dependencies are inferred from the claim text.