Media file recommendations for a search engine

1 . A computer-implemented method comprising:

determining a style preference score for a new media file based on a recommendation model, wherein the new media file is selected by a user, wherein the recommendation model is a machine learning model using a neural network; and

when the style preference score is lower than a pre-selected value:

identifying a new attribute in the new media file that is not included in the recommendation model,

updating, using a first machine learning model training step, a first coefficient in the recommendation model based on a value of the new attribute,

scoring multiple media files in a user-related search history of the user with the recommendation model based on a distance for each of the multiple media files in a multi-dimensional space defined by classifiers to form a ranked list based on the updated first coefficient, and

updating, using a second machine learning model training step, a second coefficient in the recommendation model when the new media file is not at a top position of the ranked list, relative to the multiple media files in the user-related search history; and

storing the updated recommendation model in a training database.

2 . The computer-implemented method of claim 1 , further comprising verifying that a respective search query associated with the new media file is different from one or more search queries in the user-related search history.

3 . The computer-implemented method of claim 1 , further comprising retrieving the new media file selected by the user after a pre-determined period of time has passed relative to a most recent media file in the user-related search history.

4 . The computer-implemented method of claim 1 , further comprising, when the style preference score is higher than the pre-selected value, increasing a cutoff to selected older files in the user-related search history.

5 . The computer-implemented method of claim 1 , wherein the identifying of the new attribute in the new media file comprises identifying a salient feature in the new media file.

6 . The computer-implemented method of claim 1 , further comprising:

retrieving, from the user-related search history, an unselected media file;

determining a style preference score for the unselected media file,

based on the recommendation model; and

updating a third coefficient in the recommendation model to decrease the style preference score for the unselected media file.

7 . The computer-implemented method of claim 1 , further comprising:

receiving a search query from the user; and

recommending, to the user, a top-ranked media file matching the search query based on the updated recommendation model.

8 . A non-transitory computer-readable medium storing a program, which when executed by a computer, configures the computer to:

determine a style preference score for a new media file based on a recommendation model, wherein the new media file is selected by a user, wherein the recommendation model is a machine learning model using a neural network; and

when the style preference score is lower than a pre-selected value:

identify a new attribute in the new media file that is not included in the recommendation model,

update, using a first machine learning model training step, a first coefficient in the recommendation model based on a value of the new attribute,

score multiple media files in a user-related search history of the user with the recommendation model based on a distance for each of the multiple media files in a multi-dimensional space defined by classifiers to form a ranked list based on the updated first coefficient, and

update, using a second machine learning model training step, a second coefficient in the recommendation model when the new media file is not at a top position of the ranked list, relative to the multiple media files in the user-related search history; and

store the updated recommendation model in a training database.

9 . The non-transitory computer-readable medium of claim 8 , wherein the program, when executed by the computer, further configures the computer to verify that a respective search query associated with the new media file is different from one or more search queries in the user-related search history.

10 . The non-transitory computer-readable medium of claim 8 , wherein the program, when executed by the computer, further configures the computer to retrieve the new media file selected by the user after a pre-determined period of time has passed relative to a most recent media file in the user-related search history.

11 . The non-transitory computer-readable medium of claim 8 , wherein the program, when executed by the computer, further configures the computer to, when the style preference score is higher than the pre-selected value, increase a cutoff to selected older files in the user-related search history.

12 . The non-transitory computer-readable medium of claim 8 , wherein the identifying of the new attribute in the new media file comprises identifying a salient feature in the new media file.

13 . The non-transitory computer-readable medium of claim 8 , wherein the program, when executed by the computer, further configures the computer to:

retrieve, from the user-related search history, an unselected media file;

determine a style preference score for the unselected media file,

based on the recommendation model; and

update a third coefficient in the recommendation model to decrease the style preference score for the unselected media file.

14 . The non-transitory computer-readable medium of claim 8 , wherein the program, when executed by the computer, further configures the computer to:

receive a search query from the user; and

recommend, to the user, a top-ranked media file matching the search query based on the updated recommendation model.

15 . A system comprising:

a processor; and

a non-transitory computer readable medium storing a set of instructions, which when executed by the processor, configure the system to:

determine a style preference score for a new media file based on a recommendation model, wherein the new media file is selected by a user, wherein the recommendation model is a machine learning model using a neural network; and

when the style preference score is lower than a pre-selected value:

identify a new attribute in the new media file that is not included in the recommendation model,

update, using a first machine learning model training step, a first coefficient in the recommendation model based on a value of the new attribute,

score multiple media files in a user-related search history of the user with the recommendation model based on a distance for each of the multiple media files in a multi-dimensional space defined by classifiers to form a ranked list based on the updated first coefficient, and

update, using a second machine learning model training step, a second coefficient in the recommendation model when the new media file is not at a top position of the ranked list, relative to the multiple media files in the user-related search history; and

store the updated recommendation model in a training database.

16 . The system of claim 15 , wherein the set of instructions, which when executed by the processor, further configure the system to verify that a respective search query associated with the new media file is different from one or more search queries in the user-related search history.

17 . The system of claim 15 , wherein the set of instructions, which when executed by the processor, further configure the system to retrieve the new media file selected by the user after a pre-determined period of time has passed relative to a most recent media file in the user-related search history.

18 . The system of claim 15 , wherein the set of instructions, which when executed by the processor, further configure the system to, when the style preference score is higher than the pre-selected value, increase a cutoff to selected older files in the user-related search history.

19 . The system of claim 15 , wherein the identifying of the new attribute in the new media file comprises identifying a salient feature in the new media file.

20 . The system of claim 15 , wherein the set of instructions, which when executed by the processor, further configure the system to:

retrieve, from the user-related search history, an unselected media file;

determine a style preference score for the unselected media file,

based on the recommendation model; and

update a third coefficient in the recommendation model to decrease the style preference score for the unselected media file.