Method and system for determining rank positions of content elements by a ranking system
A method and system for determining rank positions of a set of content elements by a ranking system. During a training phase of the ranking system, a plurality of datasets of previous user interactions with interfaces are acquired. For each dataset, a last-viewed content element in the interface is determined. A win score is determined for content elements that the user interacted with. A loss score is determined for content elements ranked lower than the last-viewed content element. During an in-use phase of the ranking system, a set of content elements is received. Predicted relevance scores are determined for each content element of the set of content elements. Rank positions of each content element of the set of content elements are determined based on the predicted relevance scores.
1. A method of determining rank positions, by a ranking system, of a set of content elements, the method executable by a server, the method comprising:
during a training phase of the ranking system, acquiring a plurality of datasets, each dataset comprising an indication of previous user interactions associated with an interface displaying a ranked list of content elements, and for each dataset of the plurality of datasets:
determining a last-viewed content element in the ranked list of content elements of the dataset, wherein the last-viewed content element was a last content element interacted with by a user prior to the interface associated with the dataset being abandoned;
determining, by the server, a win score for at least one content element that the user selected from the ranked list of content elements of the dataset;
determining a loss score for at least one content element ranked lower than the last-viewed content element; and
training, by the server, the ranking system to predict a relevance score for content elements based on win scores and loss scores for the content elements of the ranked list of content elements; and
during an in-use phase of the ranking system, the ranking system having been trained to rank content elements based on previous user interactions:
receiving, by the server, the set of content elements;
predicting, by the ranking system, a predicted relevance score for each content element of the set of content elements; and
determining, based on the predicted relevance score of each content element of the set of content elements, the rank positions of each content element of the set of content elements.
2. The method of claim 1 , wherein determining the loss score for the at least one content element ranked lower than the last-viewed content element comprises:
determining, for a content element ranked lower than the last-viewed content element, a number of content elements displayed between the content element and the last-viewed content element; and
determining, based on the number of content elements, the loss score for the content element.
3. The method of claim 1 , wherein determining the loss score for the at least one content element ranked lower than the last-viewed content element comprises:
determining, for a first content element ranked lower than the last-viewed content element, a first number of content elements displayed between the first content element and the last-viewed content element;
determining, for a second content element ranked lower than the last-viewed content element, a second number of content elements displayed between the second content element and the last-viewed content element;
determining, based on the first number of content elements, a first loss score for the first content element; and
determining, based on the second number of content elements, a second loss score for the second content element, wherein the second loss score is greater than the first loss score, and wherein the second number of content elements is lower than the first number of content elements.
4. The method of claim 1 , further comprising displaying a vertically arranged set of tiles, each tile comprising a content element of the set of content elements.
5. The method of claim 4 , wherein the vertically arranged set of tiles are displayed in the order of the rank positions.
6. The method of claim 1 , wherein for each dataset, each content element in the ranked list of content elements of the dataset was displayed to the user.
7. The method of claim 1 , further comprising, for each dataset, removing any content elements in the ranked list of content elements of the dataset that were not displayed to the user.
8. The method of claim 1 , wherein determining the rank positions of each content element comprises ranking the set of content elements in order from highest-scoring content element to lowest-scoring content element.
9. A system comprising:
at least one processor, and
memory storing a plurality of executable instructions which, when executed by the at least one processor, cause the system to:
during a training phase of a ranking system, acquire a plurality of datasets, each dataset comprising an indication of previous user interactions associated with an interface displaying a ranked list of content elements, and for each dataset of the plurality of datasets:
determine a last-viewed content element in the ranked list of content elements of the dataset, wherein the last-viewed content element was a last content element selected by a user prior to the interface associated with the dataset being abandoned;
determine a win score for at least one content element that the user selected from the ranked list of content elements of the dataset;
determine a loss score for at least one content element ranked lower than the last-viewed content element; and
train the ranking system to predict a relevance score for content elements based on win scores and loss scores for the content elements of the ranked list of content elements; and
during an in-use phase of the ranking system, the ranking system having been trained to rank content elements based on previous user interactions:
receive a set of content elements;
predict, by the ranking system, a predicted relevance score for each content element of the set of content elements; and
determine, based on the predicted relevance score of each content element of the set of content elements, rank positions of each content element of the set of content elements.
10. The system of claim 9 , wherein the instructions that cause the system to determine the loss score for the at least one content element comprise instructions that cause the system to:
determine, for a content element ranked lower than the last-viewed content element, a number of content elements displayed between the content element and the last-viewed content element; and
determine, based on the number of content elements, the loss score for the content element.
11. The system of claim 9 , wherein the instructions that cause the system to determine the loss score for the at least one content element comprise instructions that cause the system to:
determine, for a first content element ranked lower than the last-viewed content element, a first number of content elements displayed between the first content element and the last-viewed content element;
determine, for a second content element ranked lower than the last-viewed content element, a second number of content elements displayed between the second content element and the last-viewed content element;
determine, based on the first number of content elements, a first loss score for the first content element; and
determine, based on the second number of content elements, a second loss score for the second content element, wherein the second loss score is greater than the first loss score, and wherein the second number of content elements is lower than the first number of content elements.
12. The system of claim 9 , wherein the instructions cause the system to:
display, in an order of the rank positions, a vertically arranged set of tiles, each tile comprising a content element of the set of content elements.
13. The system of claim 9 , wherein for each dataset, each content element in the ranked list of content elements of the dataset was displayed to the user.
14. The system of claim 9 , wherein the instructions cause the system to, for each dataset, remove any content elements in the ranked list of content elements of the dataset that were not displayed to the user.
15. The system of claim 9 , wherein the instructions that cause the system to determine the rank positions of each content element comprise instructions that cause the system to rank the set of content elements in order from highest-scoring content element to lowest-scoring content element.
16. A non-transitory computer-readable medium comprising computer-readable instructions that, upon being executed by a system, cause the system to:
acquire a plurality of datasets, each dataset comprising an indication of previous user interactions associated with an interface displaying a ranked list of content elements, and for each dataset of the plurality of datasets:
determine a last-viewed content element in the ranked list of content elements of the dataset, wherein the last-viewed content element was a last content element selected by a user prior to the interface associated with the dataset being abandoned,
determine a win score for at least one content element that the user selected from the ranked list of content elements of the dataset,
determine a loss score for at least one content element ranked lower than the last-viewed content element, and
train a ranking system to predict a relevance score for content elements based on win scores and loss scores for the content elements of the ranked list of content elements;
receive a set of content elements;
predict, by the ranking system, a predicted relevance score for each content element of the set of content elements; and
determine, based on the predicted relevance score of each content element of the set of content elements, rank positions of each content element of the set of content elements.
17. The non-transitory computer-readable medium of claim 16 , wherein the instructions that cause the system to determine the loss score for the at least one content element comprise instructions that cause the system to:
determine, for a content element ranked lower than the last-viewed content element, a number of content elements displayed between the content element and the last-viewed content element; and
determine, based on the number of content elements, the loss score for the content element.
18. The non-transitory computer-readable medium of claim 16 , wherein the instructions that cause the system to determine the loss score for the at least one content element comprise instructions that cause the system to:
determine, for a first content element ranked lower than the last-viewed content element, a first number of content elements displayed between the first content element and the last-viewed content element;
determine, for a second content element ranked lower than the last-viewed content element, a second number of content elements displayed between the second content element and the last-viewed content element;
determine, based on the first number of content elements, a first loss score for the first content element; and
determine, based on the second number of content elements, a second loss score for the second content element, wherein the second loss score is greater than the first loss score, and wherein the second number of content elements is lower than the first number of content elements.
19. The non-transitory computer-readable medium of claim 16 , wherein the instructions further cause the system to display, in an order of the rank positions, a vertically arranged set of tiles, each tile comprising a content element of the set of content elements.
20. The non-transitory computer-readable medium of claim 16 , wherein the instructions cause the system to, for each dataset, remove any content elements in the ranked list of content elements of the dataset that were not displayed to the user.