System and method for determining item labels based on item images
A method including automatically determining, by a machine learning model trained based at least in part on sample items stored in a sample database, a query embedding vector for a query image of a query item. The method further can include determining, based on a respective embedding distance between the query image of the query item and a respective image of each of the sample items, neighboring items from among the sample items. The respective embedding distance can be calculated based on the query embedding vector for the query image and a respective embedding vector for the respective image of each of the sample items. Each of the sample items can include the respective image and at least one respective item label. The method also can include determining a respective normalized weight for each of the neighboring items based on the respective embedding distance between the query image and the respective image of the each of the neighboring items. The method additionally can include determining a query item label of the query item based on a weighted majority vote by the neighboring items via the respective normalized weight for the each of the neighboring items. The method further can include upon determining that the query item label of the query item is different from a first item label of the query item. storing the query item with the query item label in a product database. The method also can include selectively updating the sample items stored in the sample database from items in the product database. In addition, the method can include re-training the machine learning model based at least in part on the sample items in the sample database, as updated. Other embodiments are disclosed.
1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing computing instructions that, when executed on the one or more processors, cause the one or more processors to perform:
automatically determining, by a machine learning model trained based at least in part on sample items stored in a sample database, a query embedding vector for a query image of a query item;
determining, based on a respective embedding distance between the query image of the query item and a respective image of each of the sample items, neighboring items from among the sample items, wherein:
the respective embedding distance is calculated based on the query embedding vector for the query image and a respective embedding vector for the respective image of each of the sample items; and
each of the sample items comprises the respective image and at least one respective item label;
determining a respective normalized weight for each of the neighboring items based on the respective embedding distance between the query image and the respective image of the each of the neighboring items;
determining a query item label of the query item based on a weighted majority vote by the neighboring items via the respective normalized weight for the each of the neighboring items;
upon determining that the query item label of the query item is different from a first item label of the query item, storing the query item with the query item label in a product database;
selectively updating the sample items stored in the sample database from items in the product database; and
re-training the machine learning model based at least in part on the sample items in the sample database, as updated.
2. The system in claim 1 , wherein:
the computing instructions, when executed on the one or more processors, further cause the one or more processors to perform:
before retrieving the sample items:
selectively determining a new sample image associated with at least one new image label for a new sample item of the sample items, wherein:
the respective image of the new sample item comprises the new sample image; and
the at least one respective item label of the new sample item comprises the at least one new image label;
determining the respective embedding vector for the new sample item; and
storing the new sample item in the sample database.
3. The system in claim 1 , wherein:
determining the neighboring items further comprises using a k-nearest neighbors (K-NN) search engine to search for the neighboring items for the query item.
4. The system in claim 1 , wherein:
the respective embedding distance between the query image and the respective image of the each of the neighboring items is a Euclidean distance between the query embedding vector for the query image and the respective embedding vector of the each of the neighboring items.
5. The system in claim 1 , wherein:
the machine learning model further comprises a neural network model.
6. The system in claim 5 , wherein:
the computing instructions, when executed on the one or more processors, further cause the one or more processors to perform:
before determining automatically the query embedding vector, training the neural network model based at least in part on the respective image and the at least one respective item label of the each of the sample items.
7. The system in claim 1 , wherein:
determining the respective normalized weight for the each of the neighboring items further comprises calculating the respective normalized weight by:
exp
(
d
i
)
Σ
i
exp
(
d
i
)
,
∀
i
∈
[
K
]
wherein:
d i : the respective embedding distance between the query image and its i-th neighbor; and
K: the neighboring items.
8. The system in claim 1 , wherein:
determining the query item label of the query item based on the weighted majority vote further comprises:
tallying at least one respective weighted group vote from each group of one or more groups of the neighboring items by summing up the respective normalized weight for each of one or more respective items of the each group of the one or more groups, wherein:
the neighboring items comprise the one or more respective items; and
the at least one respective item label of the each of the one or more respective items is identical within the each group; and
determining the query item label of the query item to be the at least one respective item label of each of the one or more respective items of a wining group of the one or more groups based on the at least one respective weighted group vote from the wining group.
9. The system in claim 1 , wherein:
the computing instructions, when executed on the one or more processors, further cause the one or more processors to perform:
upon determining that a designation error exists and before storing the query item in the product database, transmitting an alert to a user to be displayed on a user interface executed on a user device for the user based at least in part on the designation error.
10. The system in claim 9 , wherein:
the computing instructions, when executed on the one or more processors, further cause the one or more processors to perform:
before transmitting the alert to the user, determining a confidence level status based on whether a confidence level for the query item label is at least as great as a predetermined threshold; and
transmitting the alert to the user further comprises transmitting the alert to the user further based on the confidence level status.
11. A method being implemented via execution of computing instructions configured to run at-one or more processors and stored at one or more non-transitory computer-readable media, the method comprising:
automatically determining, by a machine learning model trained based at least in part on sample items stored in a sample database, a query embedding vector for a query image of a query item;
determining, based on a respective embedding distance between the query image of the query item and a respective image of each of the sample items, neighboring items from among the sample items, wherein:
the respective embedding distance is calculated based on the query embedding vector for the query image and a respective embedding vector for the respective image of each of the sample items; and
each of the sample items comprises the respective image and at least one respective item label;
determining a respective normalized weight for each of the neighboring items based on the respective embedding distance between the query image and the respective image of the each of the neighboring items;
determining a query item label of the query item based on a weighted majority vote by the neighboring items via the respective normalized weight for the each of the neighboring items;
upon determining that the query item label of the query item is different from a first item label of the query item, storing the query item with the query item label in a product database;
selectively updating the sample items stored in the sample database from items in the product database; and
re-training the machine learning model based at least in part on the sample items in the sample database, as updated.
12. The method in claim 11 , further comprising:
before retrieving the sample items:
selectively determining a new sample image associated with at least one new image label for a new sample item of the sample items, wherein:
the respective image of the new sample item comprises the new sample image; and
the at least one respective item label of the new sample item comprises the at least one new image label;
determining the respective embedding vector for the new sample item; and
storing the new sample item in the sample database.
13. The method in claim 11 , wherein:
determining the neighboring items further comprises using a k-nearest neighbors (K-NN) search engine to search for the neighboring items for the query item.
14. The method in claim 11 , wherein:
the respective embedding distance between the query image and the respective image of the each of the neighboring items is a Euclidean distance between the query embedding vector for the query image and the respective embedding vector of the each of the neighboring items.
15. The method in claim 11 , wherein:
the machine learning model further comprises a neural network model.
16. The method in claim 15 , further comprising, before determining automatically the query embedding vector:
training the neural network model based at least in part on the respective image and the at least one respective item label of the each of the sample items.
17. The method in claim 11 , wherein:
determining the respective normalized weight for the each of the neighboring items further comprises calculating the respective normalized weight by:
exp
(
d
i
)
Σ
i
exp
(
d
i
)
,
∀
i
∈
[
K
]
wherein:
d i : the respective embedding distance between the query image and its i-th neighbor; and
K: the neighboring items.
18. The method in claim 11 , wherein:
determining the query item label of the query item based on the weighted majority vote further comprises:
tallying at least one respective weighted group vote from each group of one or more groups of the neighboring items by summing up the respective normalized weight for each of one or more respective items of the each group of the one or more groups, wherein:
the neighboring items comprise the one or more respective items; and
the at least one respective item label of the each of the one or more respective items is identical within the each group; and
determining the query item label of the query item to be the at least one respective item label of each of the one or more respective items of a wining group of the one or more groups based on the at least one respective weighted group vote from the wining group.
19. The method in claim 11 , further comprising:
upon determining that a designation error exists and before storing the query item in the product database, transmitting an alert to a user to be displayed on a user interface executed on a user device for the user based at least in part on the designation error.
20. The method in claim 19 , further comprising:
before transmitting the alert to the user, determining a confidence level status based on whether a confidence level for the query item label is at least as great as a predetermined threshold,
wherein:
transmitting the alert to the user further comprises transmitting the alert to the user further based on the confidence level status.