Offering automobile recommendations from generic features learned from natural language inputs

Patent metadata

Field	Value
Publication number	US-11915293-B2
Application number	US-202017094088-A
Country	US
Kind code	B2
Filing date	Nov 10, 2020
Priority date	Jan 22, 2019
Publication date	Feb 27, 2024
Grant date	Feb 27, 2024

Abstract

Various embodiments are generally directed to techniques to provide specific vehicle recommendations to generic user requests. A method for providing the specific vehicle recommendation includes: receiving a generic automobile request from a user, applying a machine learning model (MLM) trained by a corpus of reviews to the received request, and generating, by the MLM, a recommendation for at least one specific automobile feature based on the generic automobile request.

First claim

What is claimed is:

1. A method, comprising:
receiving a first data set and a second data set from a corpus of one or more expert automobile reviews, the first data set and the second data set comprising generic text related to a plurality of automobile makes and models and specific text related to at least one feature of at least one of the plurality of automobile makes and models, wherein the generic text of the corpus of one or more expert automobile reviews is related to the specific text of the corpus of the one or more expert automobile reviews;
training, by at least one computer processor, a machine learning model (MLM) based on at least the first data set;
generating, by the MLM, a respective probability distribution for one or more specific automobile makes and models in relation to generic automobile text by analyzing a relationship between the generic text of the corpus of one or more expert automobile reviews and the specific text of the corpus of one or more expert automobile reviews;
receiving a third data set comprising a generic automobile text, the generic automobile text comprising a preference of an account; and
generating, by the MLM, a recommendation comprising a specific automobile make and model corresponding to the generic automobile text of the third data set, the recommendation based on the probability distribution for the specific automobile make and model.

2. The method of claim 1 , further comprising:
pre-processing the first data set and the second data set to: (i) include a plurality of sentences from the corpus of expert reviews, and (ii) remove predetermined verbs, pronouns, and stop words from the plurality of sentences, wherein the MLM is an embedded MLM, and wherein the pre-processed data is processed by a sentence encoder that is part of the MLM.

3. The method of claim 1 , wherein the MLM is a transfer learning model trained by a first use type, wherein the first use type is associated with the first data set and the second data set, wherein the first use type is reused on the third data set, wherein the third data set is associated with a second use type that is distinct from the first use type, the method further comprising:
updating the MLM based on an adjustment to the probability distribution for the specific automobile make and model as a result of input specifying a preference for the specific automobile make and model.

4. The method of claim 3 , wherein the second use type is based on a request from the account, and wherein the recommendation comprising the specific automobile make and model corresponding to the generic automobile text of the third data set relates to an automobile preference of the account.

5. The method of claim 1 , further comprising:
receiving a second request from another account, the second request including a generic automobile preference; and
generating a second automobile suggestion responsive to the second request based on the second request and using the MLM, the second automobile suggestion including an automobile make and model.

6. The method of claim 5 , further comprising:
providing an interface for receiving the generic automobile preference, wherein the interface includes a single field for entering an entirety of the generic automobile preference, and wherein the generic automobile preference consists solely of generic language; and
displaying the second automobile suggestion on a display of a computer device.

7. The method of claim 1 , further comprising:
pre-processing the first data set and second data set to: (i) include a plurality of sentences from the corpus of expert reviews, and (ii) remove predetermined verbs, pronouns, and stop words from the plurality of sentences, and wherein the pre-processed data is then processed by a word frequency-based sentence vectorizer that is part of the MLM.

8. A non-transitory computer-readable storage medium storing computer-readable program code executable by a processor to:
receive a first data set and a second data set from a corpus of one or more expert automobile reviews, the first data set and the second data set comprising generic text related to a plurality of automobile makes and models and specific text related to at least one feature of at least one of the plurality of automobile makes and models, wherein the generic text of the corpus of one or more expert automobile reviews is related to the specific text of the corpus of the one or more expert automobile reviews;
train, by at least one computer processor, a machine learning model (MLM) based on at least the first data set;
generate, by the MLM, a respective probability distribution for one or more specific automobile makes and models in relation to generic automobile text by analyzing a relationship between the generic text of the corpus of one or more expert automobile reviews and the specific text of the corpus of one or more expert automobile reviews;
receive a third data set comprising a generic automobile text, the generic automobile text comprising a preference of an account; and
generate, by the MLM, a recommendation comprising a specific automobile make and model corresponding to the generic automobile text of the third data set, the recommendation based on the probability distribution for the specific automobile make and model.

9. The non-transitory computer-readable storage medium of claim 8 , storing computer-readable program code executable by the processor to:
pre-process the first data set and the second data set to: (i) include a plurality of sentences from the corpus of expert reviews, and (ii) remove predetermined verbs, pronouns, and stop words from the plurality of sentences, wherein the MLM is an embedded MLM, and wherein the pre-processed data is processed by a sentence encoder that is part of the MLM.

10. The non-transitory computer-readable storage medium of claim 8 , wherein the MLM is a transfer learning model trained by a first use type, wherein the first use type is associated with the first data set and the second data set, wherein the first use type is reused on the third data set, wherein the third data set is associated with a second use type that is distinct from the first use type, the medium storing computer-readable program code executable by the processor to:
update the MLM based on an adjustment to the probability distribution for the specific automobile make and model as a result of input specifying a preference for the specific automobile make and model.

11. The non-transitory computer-readable storage medium of claim 10 , wherein the second use type is based on a request from the account, and wherein the recommendation comprising the specific automobile make and model corresponding to the generic automobile text of the third data set relates to an automobile preference of the account.

12. The non-transitory computer-readable storage medium of claim 8 , storing computer-readable program code executable by the processor to:
receive a second request from another account, the second request including a generic automobile preference; and
generate a second automobile suggestion responsive to the second request based on the second request and using the MLM, the second automobile suggestion including an automobile make and model.

13. The non-transitory computer-readable storage medium of claim 12 , storing computer-readable program code executable by the processor to:
provide an interface for receiving the generic automobile preference, wherein the interface includes a single field for entering an entirety of the generic automobile preference, and wherein the generic automobile preference consists solely of generic langu

Assignees

• Capital One Services Llc

Inventors

• Price Micah
• Wylie Stephen
• Hooshmand Habeeb
• Hoover Jason
• Dagley Geoffrey
• Tang Qiaochu

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06Q30/0631 Primary

  Recommending goods or services · CPC title

• G06F16/9538

  Presentation of query results · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N20/00

  Machine learning · CPC title