Ranking product search results

1. A system, comprising:

one or more processors configured to:

determine a plurality of sample pairs corresponding to a query word, wherein a sample pair of the plurality of sample pairs is associated with a first product and a second product responsive to the query word, wherein one of the first product and the second product is assigned a higher ranking priority than the other one of the first product and the second product based at least in part on historical user action data associated with the query word, wherein to determine the plurality of sample pairs corresponding to the query word includes to:

determine a plurality of products responsive to the query word based at least in part on the historical user action data associated with the query word;

determine a historical metric value for each product of the plurality of products based at least in part on the historical user action data associated with the query word;

determine a plurality of pairs of products from the plurality of products;

determine the plurality of sample pairs from the plurality of pairs of products, wherein each of the plurality of sample pairs meets a condition associated with the historical metric value; and

assign a higher priority to a product in each sample pair of the plurality of sample pairs based at least in part on respective historical metric values corresponding to the two products of the sample pair;

determine a plurality of feature score values corresponding to a set of ranking features associated with the query word for each product associated with each of at least a subset of the plurality of sample pairs;

determine a set of weightings corresponding to the set of ranking features based at least in part on the plurality of feature score values and ranking priority assignments corresponding to the at least subset of the plurality of sample pairs, wherein to determine the set of weightings corresponding to the set of ranking features includes to train a machine learning model based at least in part on the plurality of feature score values and indications of higher ranking priority or lower ranking priority corresponding to the two products of each sample pair of the plurality of sample pairs;

receive a subsequent search request including the query word;

rank at least the plurality of products responsive to the query word based at least in part on the set of weightings; and

output one or more ranked product search results; and

one or more memories coupled to the one or more processors and configured to provide instructions to the one or more processors.

2. The system of claim 1 , wherein to rank the at least plurality of products includes to:

determine, for a product of the plurality of products, a corresponding feature score value for each ranking feature of the set of ranking features;

determine a ranking score for the product based at least in part on the corresponding feature score values and the set of weightings; and

rank the product among the plurality of products based at least in part on the ranking score.

3. The system of claim 1 , wherein the historical metric values comprise one or more of the following: a conversion rate and a click-through rate.

4. The system of claim 1 , wherein to determine the plurality of feature score values corresponding to the set of ranking features associated with the query word corresponding to each product associated with each of the at least subset of the plurality of sample pairs includes to:

extract a set of attribute values corresponding to the set of ranking features associated with the product; and

compute the plurality of feature score values for the product based at least in part on the set of attribute values.

5. The system of claim 1 , wherein the set of ranking features associated with the query word is associated with a category to which the query word belongs.

6. The system of claim 1 , wherein the historical user action data includes one or more of the following: access logs, purchase logs, or selection logs.

7. A method, comprising:

determining a plurality of sample pairs corresponding to a query word, wherein a sample pair of the plurality of sample pairs is associated with a first product and a second product responsive to the query word, wherein one of the first product and the second product is assigned a higher ranking priority than the other one of the first product and the second product based at least in part on historical user action data associated with the query word, wherein determining the plurality of sample pairs corresponding to the query word includes:

determining a plurality of products responsive to the query word based at least in part on the historical user action data associated with the query word;

determining a historical metric value for each product of the plurality of products based at least in part on the historical user action data associated with the query word;

determining a plurality of pairs of products from the plurality of products;

determining the plurality of sample pairs from the plurality of pairs of products, wherein each of the plurality of sample pairs meets a condition associated with the historical metric value; and

assigning a higher priority to a product in each sample pair of the plurality of sample pairs based at least in part on respective historical metric values corresponding to the two products of the sample pair;

determining a plurality of feature score values corresponding to a set of ranking features associated with the query word for each product associated with each of at least a subset of the plurality of sample pairs;

determining a set of weightings corresponding to the set of ranking features based at least in part on the plurality of feature score values and ranking priority assignments corresponding to the at least subset of the plurality of sample pairs, wherein determining the set of weightings corresponding to the set of ranking features includes training a machine learning model based at least in part on the plurality of feature score values and indications of higher ranking priority or lower ranking priority corresponding to the two products of each sample pair of the plurality of sample pairs;

receiving a subsequent search request including the query word;

ranking at least the plurality of products responsive to the query word based at least in part on the set of weightings; and

outputting one or more ranked product search results.

8. The method of claim 7 , wherein ranking the at least plurality of products includes:

determining, for a product of the plurality of products, a corresponding feature score value for each ranking feature of the set of ranking features;

determining a ranking score for the product based at least in part on the corresponding feature score values and the set of weightings; and

ranking the product among the plurality of products based at least in part on the ranking score.

9. The method of claim 7 , wherein the historical metric values comprise one or more of the following: a conversion rate and a click-through rate.

10. The method of claim 7 , wherein determining the plurality of feature score values corresponding to the set of ranking features associated with the query word corresponding to each product associated with each of the at least subset of the plurality of sample pairs includes:

extracting a set of attribute values corresponding to the set of ranking features associated with the product; and

computing the plurality of feature score values for the product based at least in part on the set of attribute values.

11. The method of claim 7 , wherein the set of ranking features associated with the query word is associated with a category to which the query word belongs.

12. The method of claim 7 , wherein the historical user action data includes one or more of the following: access logs, purchase logs, or selection logs.

13. A computer program product, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

determining a plurality of sample pairs corresponding to a query word, wherein a sample pair of the plurality of sample pairs is associated with a first product and a second product responsive to the query word, wherein one of the first product and the second product is assigned a higher ranking priority than the other one of the first product and the second product based at least in part on historical user action data associated with the query word, wherein determining the plurality of sample pairs corresponding to the query word includes:

determining a plurality of products responsive to the query word based at least in part on the historical user action data associated with the query word;

determining a historical metric value for each product of the plurality of products based at least in part on the historical user action data associated with the query word;

determining a plurality of pairs of products from the plurality of products;

determining the plurality of sample pairs from the plurality of pairs of products, wherein each of the plurality of sample pairs meets a condition associated with the historical metric value; and

assigning a higher priority to a product in each sample pair of the plurality of sample pairs based at least in part on respective historical metric values corresponding to the two products of the sample pair;

determining a plurality of feature score values corresponding to a set of ranking features associated with the query word for each product associated with each of at least a subset of the plurality of sample pairs;

determining a set of weightings corresponding to the set of ranking features based at least in part on the plurality of feature score values and ranking priority assignments corresponding to the at least subset of the plurality of sample pairs, wherein determining the set of weightings corresponding to the set of ranking features includes training a machine learning model based at least in part on the plurality of feature score values and indications of higher ranking priority or lower ranking priority corresponding to the two products of each sample pair of the plurality of sample pairs;

receiving a subsequent search request including the query word;

ranking at least the plurality of products responsive to the query word based at least in part on the set of weightings; and

outputting one or more ranked product search results.