Item-to-item recommendations

1. A system comprising:

a memory having instructions stored thereon, and a processor configured to read the instructions to:

extract, by the processor, feature data for a plurality of items, wherein the feature data comprises at least one of attribute data, content data, and image data;

embed the image data of the plurality of items to reduce dimensionality of the image data based on at least one of a deep learning analysis or a latent semantic analysis;

convert the feature data for the plurality of items into feature vectors for the plurality of items, wherein the feature vectors comprise at least one of an attribute vector, a content vector, and an image vector;

determine a degree of similarity between each of the plurality of items based, at least in part, on a distance between each of the feature vectors for the plurality of items;

generate, by the processor, one or more weights for each of the feature vectors based, at least in part, on a logistic regression model;

receive data for an item of a click stream;

generate data for a recommended candidate item for the data of the received item, based on image data similarity between a candidate item corresponding to the recommended candidate item data and the data of the received item;

generate feature data of the recommended candidate item data and the data of the received item, the feature data comprising one or more similarities between the data of the received item and data corresponding to a respective recommended candidate item, wherein the feature data of the data of the received item and the recommended candidate item data are converted into a received item vector and a recommended candidate item vector, respectively;

determine a degree of similarity between the data of the received item and the recommended candidate item data based on a distance between the received item vector and the recommended candidate item vector;

generate ranked recommended candidate item data based on the received item vector, the recommended candidate item vector, and the generated one or more weights for each of the feature vectors determined for candidate test item data, wherein the candidate test item data comprises data corresponding to the recommended candidate item data; and

output the generated ranked recommended candidate item data wherein the generated ranked recommended candidate item data is based on a last received item within the click stream.

2. The system of claim 1 , wherein the item is a cold start item and the processor is further configured to read the instructions to generate, based on the image data similarity between the candidate item and the received cold start item, the recommended candidate item data of the received cold start item.

3. The system of claim 1 , wherein the processor is further configured to generate the recommended candidate items based on content data similarity between the candidate item and the received item.

4. The system of claim 1 , wherein the processor is further configured to:

receive data for at least another item in a sequential order and generate a click stream for the received item and the at least another item; and

determine averaged feature data of the received item and the at least another item within the click stream.

5. The system of claim 4 , wherein the processor is further configured to:

generate recommended candidate items for the at least another item within the click stream;

generate feature data of the averaged feature data and the recommended candidate items for the at least another item; and

rank the recommended candidate items for the at least another item.

6. The system of claim 1 , wherein the candidate test item data comprises one or more of image data of the recommended candidate item data, content data of the recommended candidate item data, price data of the recommended candidate item data, and categorical data of the recommended candidate item data.

7. The system of claim 6 , wherein the processor is further configured to:

generate data for candidate test pairs of items based on click stream training data;

generate feature data for the data for the candidate pairs of items; and

determine at least one weight for the feature data of the candidate test item data.

8. A method comprising:

extracting, by at least one processor, feature data for a plurality of items, wherein the feature data comprises at least one of attribute data, content data, and image data;

embedding the image data of the plurality of items to reduce dimensionality of the image data based on at least one of a deep learning analysis or a latent semantic analysis;

converting, by the at least one processor, the feature data for the plurality of items into feature vectors for the plurality of items, wherein the feature vectors comprise at least one of an attribute vector, a content vector, and an image vector;

determining, by the at least one processor, a degree of similarity between each of the plurality of items based, at least in part, on a distance between each of the feature vectors for the plurality of items;

generating, by the at least one processor, one or more weights for each of the feature vectors based, at least in part, on a logistic regression model;

receiving, by the at least one processor, data for an item of a click stream;

generating, by the at least one processor, data for a recommended candidate item for the data of the received item, based on image data similarity between a candidate item corresponding to the recommended candidate item data and the data of the received item;

generating, by the at least one processor, feature data of the recommended candidate item data and the data of the received item, the feature data comprising one or more similarities between the data of the received item and data corresponding to a respective recommended candidate item, wherein the feature data of the data of the received item and the recommended candidate item data are converted into a received item vector and a recommended candidate item vector, respectively;

determining, by the at least one processor, a degree of similarity between the data of the received item and the recommended candidate item data based on a distance between the received item vector and the recommended candidate item vector;

generating, by the at least one processor, ranked recommended candidate item data based on the received item vector, the recommended candidate item vector, and the generated one or more weights for each of the feature vectors determined for candidate test item data, wherein the candidate test item data comprises data corresponding to the recommended candidate item data; and

outputting the generated ranked recommended candidate item data, wherein the generated ranked recommended candidate item data is based on a last received item within the click stream.

9. The method of claim 8 , wherein the item is a cold start item, and generating the recommended candidate item data comprises generating, based on the image data similarity between the candidate item and the received cold start item, the recommended candidate item data of the received cold start item.

10. The method of claim 8 , wherein generating the recommended candidate items comprises generating the recommended candidate items based on content data similarity between the candidate item and the received item.

11. The method of claim 8 , further comprising:

receiving, by the at least one processor, data for at least another item in a sequential order and generating, by the at least one processor, a click stream for the received item and the at least another item; and

determining, by the at least one processor, averaged feature data of the received item and the at least another item within the click stream.

12. The method of claim 11 , further comprising:

generating, by the at least one processor, recommended candidate items for the at least another item within the click stream;

generating, by the at least one processor, feature data of the averaged feature data and the recommended candidate items for the at least another; and

ranking, by the at least one processor, the recommended candidate items for the at least another item.

13. The method of claim 8 , wherein the candidate test item data comprises one or more of image data of the recommended candidate item data, content data of the recommended candidate item data, price data of the recommended candidate item data, and categorical data of the recommended candidate item data.

14. The method of claim 13 , further comprising:

generating, by the at least one processor, data for candidate test pairs of items based on click stream training data;

generating, by the at least one processor, feature data for the data for the candidate pairs of items; and

determining, by the at least one processor, at least one weight for the feature data of the candidate test item data.

15. A computer program product comprising:

a non-transitory computer readable medium having program instructions stored thereon, the program instructions executable by one or more processors, the program instructions comprising:

extracting, by the one or more processors, feature data for a plurality of items, wherein the feature data comprises at least one of attribute data, content data, and image data;

embedding the image data of the plurality of items to reduce dimensionality of the image data based on at least one of a deep learning analysis or a latent semantic analysis;

converting, by the one or more processors, the feature data for the plurality of items into feature vectors for the plurality of items, wherein the feature vectors comprise at least one of an attribute vector, a content vector, and an image vector;

determining, by the one or more processors, a degree of similarity between each of the plurality of items based, at least in part, on a distance between each of the feature vectors for the plurality of items;

generating, by the one or more processors, one or more weights for each of the feature vectors based, at least in part, on a logistic regression model;

receiving data for an item of a click stream;

generating data for a recommended candidate item for the data of the received item, based on image data similarity between a candidate item corresponding to the recommended candidate item data and the data of the received item;

generating feature data of the recommended candidate item data and the data of the received item, the feature data comprising one or more similarities between the data of the received item and data corresponding to a respective recommended candidate item, wherein the feature data of the data of the received item and the recommended candidate item data are converted into a received item vector and a recommended candidate item vector, respectively;

determining, by the one or more processors, a degree of similarity between the data of the received item and the recommended candidate item data based on a distance between the received item vector and the recommended candidate item vector;

generating ranked recommended candidate item data based on the data and the generated one or more weights for each of the feature vectors determined for candidate test item data, wherein the candidate test item data comprises data corresponding to the recommended candidate item data; and

outputting the generated ranked recommended candidate item data, wherein the generated ranked recommended candidate item data is based on a last received item within the click stream.

16. The computer program product of claim 15 , wherein:

the item is a cold start item; and

generating the recommended candidate items comprises generating, based on the image data similarity between the candidate item and the received cold start item and content data similarity between the candidate item and the received item, the recommended candidate item data of the received cold start item.

17. The computer program product of claim 15 , wherein the program instructions further comprise:

receiving data for at least another item in a sequential order and generating a click stream for the received item and the at least another item; and

determining averaged feature data of the received item and the at least another item within the click stream.

18. The computer program product of claim 17 , wherein the program instructions further comprise:

generating recommended candidate items for the at least another item within the click stream;

generating feature data of the averaged feature data and the recommended candidate items for the at least another; and

ranking the recommended candidate items for the at least another item.

19. The computer program product of claim 15 , wherein the candidate test item data comprises one or more of image data of the recommended candidate item data, content data of the recommended candidate item data, price data of the recommended candidate item data, and categorical data of the recommended candidate item data.

20. The computer program product of claim 19 , wherein the program instructions further comprise:

generating data for candidate test pairs of items based on click stream training data;

generating feature data for the data for the candidate pairs of items; and

determining at least one weight for the feature data of the candidate test item data.