US 12,769,945 B2Granted

Systems and methods for classifying token sequence embeddings

AssigneeCapital One Services, LLC

InventorsSamuel Sharpe, Brian Barr, Jeremy Goodsitt

Life of this patent

Estimating the term…

Loading patent…

Abstract

Systems and methods for classifying token sequences using clustering. In some aspects, the system receives a first set of input sequences and generates a corresponding first plurality of embeddings using a first machine learning model. The system generate a plurality of clusters, each containing embeddings, using a clustering model. For each cluster in the plurality of clusters, the system generates a set of common components in the cluster. The system generates a set of universal components based on the plurality of clusters. For each cluster in the plurality of clusters, the system generates a set of unique components by removing the set of universal components from its associated set of common components. The system classifies a second plurality of embeddings using the sets of unique components.

Claims18

Dependent claims

A system for improving model performance by reducing vector search spaces, the system comprising one or more processors and one or more non-transitory, computer-readable media comprising instructions that, when executed, cause operations comprising:

obtaining, during a communication session, first data comprising one or more first inputs and second data comprising a query;
providing a first machine learning model with the one or more first inputs to obtain an earlier plurality of embeddings and with the query to obtain a later plurality of embeddings;
clustering the earlier plurality of embeddings using a clustering model to generate clusters;
generating universal vectors by ranking real-value segments of the clusters according to occurrence frequencies, wherein each vector of the universal vectors comprises a real-value segment associated with an occurrence frequency greater than an inter-cluster threshold;
generating, for each cluster in the clusters, a set of common vectors in the cluster by selecting one or more vectors of the cluster having an intra-cluster occurrence frequency that is greater than an intra-cluster admittance threshold;
generating, for each cluster in the clusters, a set of unique vectors by removing the universal vectors from the set of common vectors;
generating a set of archetype vectors by (i) determining a set of unique vectors by removing the universal vectors from the set of common vectors and (ii) determining the set of archetype vectors based on the set of unique vectors without using vectors of the universal vectors;
generating distance metrics between the later plurality of embeddings and the set of archetype vectors;
classifying the later plurality of embeddings with a classification associated with a minimum distance of the distance metrics;
and
presenting, in a user interface, a target response during the communication session by generating the target response based on the classification.

8 more elements

A method for improving model performance by reducing vector search spaces, comprising:

obtaining, during one or more communication sessions, first data comprising one or more first inputs and second data comprising a query;
providing a first machine learning model with the one or more first inputs to obtain an earlier plurality of embeddings and with the query to obtain a later-a first plurality of embeddings;
clustering the earlier plurality of embeddings using a clustering model to generate clusters;
generating universal vectors by ranking real-value segments of the clusters according to occurrence frequencies, wherein each vector of the universal vectors comprises a real-value segment associated with an occurrence frequency greater than an inter-cluster threshold;
generating, for each cluster in the clusters, a set of common vectors in the cluster by selecting one or more vectors of the cluster having an intra-cluster occurrence frequency that is greater than an intra-cluster admittance threshold;
generating a set of archetype vectors by (i) determining a set of unique vectors by removing the universal vectors from the set of common vectors and (ii) determining the set of archetype vectors based on the set of unique vectors without using vectors of the universal vectors;
generating distance metrics between the later plurality of embeddings and the set of archetype vectors;
classifying the later plurality of embeddings with a classification associated with a minimum distance of the distance metrics;
and
presenting, in a user interface that is used during at least one session of the one or more communication sessions, a target response based on the classification.

7 more elements

One or more non-transitory computer-readable media comprising instructions that, when executed by one or more processors, cause operations comprising:

obtaining, during one or more communication sessions, first data comprising one or more first inputs and second data comprising a query;
providing a first machine learning model with the one or more first inputs to obtain a first plurality of embeddings and with the query to obtain a second plurality of embeddings;
clustering the first plurality of embeddings using a clustering model to generate clusters;
generating universal vectors by ranking real-value segments of the clusters according to occurrence frequencies, wherein each vector of the universal vectors comprises a real-value segment associated with an occurrence frequency greater than an inter-cluster threshold;
generating, for each cluster in the clusters, a set of common vectors in the cluster by selecting one or more vectors of the cluster having an intra-cluster occurrence frequency that is greater than an intra-cluster admittance threshold;
generating a set of archetype vectors by (i) determining a set of unique vectors by removing the universal vectors from the set of common vectors and (ii) determining the set of archetype vectors based on the set of unique vectors;
without using vectors of the universal vectors;
generating distance metrics between the second plurality of embeddings and the set of archetype vectors;
classifying the second plurality of embeddings with a classification associated with a minimum distance of the distance metrics;
and
presenting, in a user interface that is used during at least one session of the one or more communication sessions, a target response based on the classification.

8 more elements

Dependencies are inferred from the claim text.