E-commerce recommendation system and method

Patent metadata

Field	Value
Publication number	US-2016110794-A1
Application number	US-201414518431-A
Country	US
Kind code	A1
Filing date	Oct 20, 2014
Priority date	Oct 20, 2014
Publication date	Apr 21, 2016
Grant date	—

Abstract

Disclosed herein is item recommender that uses a model trained using a combination of at least visual item similarity training data and social activity training data. The model may be used, for example, to identify a set of recommended products having similar visual features as a given product. The set of recommended products may be presented to the user along with the given product. The model may be continuously updated using feedback from users to identify the features considered to be important to the users relative to other features.

First claim

1 . A method comprising:
determining, by at least one computing device, visual similarity training data for a plurality of items, the visual similarity training data identifying, for each pair of items of the plurality of items, a level of visual similarity between images of the items in the pair; determining, by the at least one computing device, social activity training data for the plurality of items, the social activity training data identifying, for each pair of items of the plurality of items, an indicator of whether a shared user interest in the pair of items exists; training, by the at least one computing device, a model using a training data comprising the image similarity training data and the social activity training data; and generating, by the at least one computing device, a set of recommended items using the trained model.

2 . The method of claim 1 , determining social activity training data further comprising determine the social training data using observed user behavior comprising one or more of a co-view, co-purchase and co-favorite observed user behavior.

3 . The method of claim 1 , the determining the image similarity training data further comprising:
extracting visual characteristics from each item's image, the extracting comprising:
making a determination whether the item's image contains one or more extraneous objects other than the item; and
selecting an extraction tool based on the determination.

4 . The method of claim 3 , the selecting further comprising:
selecting an edge-based background removal extraction tool if the determination indicates an absence of the one or more extraneous objects; and selecting a deformable part-based model to extract the item's image if the determination indicates a presence of one or more extraneous objects.

5 . The method of claim 1 , further comprising:
receiving, by the at least one computing device, user feedback comprising information identifying user item selection; and retraining, by the at least one computing device, the model using the training data comprising the image similarity training data and the social activity training data and the user feedback, the retraining comprising identifying, using the user feedback, a weighting for at least the image similarity training data and the social activity training data to be used in retraining the model.

6 . The method of claim 5 , the training data further comprising item metadata.

7 . The method of claim 1 , generating a set of recommended items using the trained model further comprising:
determining, using the trained model, a click prediction for each of the items of the plurality, each item's click prediction is based at least on the item's features, the item's features comprising at least visual similarity and social activity features; and selecting one or more items of the plurality based on each item's click prediction.

8 . A system comprising:
at least one computing device, each computing device comprising a processor and a storage medium for tangibly storing thereon program logic for execution by the processor, the stored program logic comprising: first determining logic for determining visual similarity training data for a plurality of items, the visual similarity training data identifying, for each pair of items of the plurality of items, a level of visual similarity between images of the items in the pair; second determining logic for determining social activity training data for the plurality of items, the social activity training data identifying, for each pair of items of the plurality of items, an indicator of whether a shared user interest in the pair of items exists; training logic for training a model using a training data comprising the image similarity training data and the social activity training data; and generating logic for generating, using the trained model, a set of recommended items.

9 . The system of claim 8 , the second determining logic for determining social activity training data further comprising determining logic for determining the social training data using observed user behavior comprising one or more of a co-view, co-purchase and co-favorite observed user behavior.

10 . The system of claim 8 , the first determining logic for determining the image similarity training data further comprising:
extracting logic for extracting visual characteristics from each item's image, the extracting logic comprising:
determining logic for making a determination whether the item's image contains one or more extraneous objects other than the item; and
selecting logic for selecting an extraction tool based on the determination.

11 . The system of claim 10 , the selecting logic further comprising:
first selecting logic for selecting an edge-based background removal extraction tool if the determination indicates an absence of the one or more extraneous objects; and second selecting logic for selecting a deformable part-based model to extract the item's image if the determination indicates a presence of one or more extraneous objects.

12 . The system of claim 8 , the stored program logic further comprising:
receiving logic for receiving user feedback comprising information identifying user item selection; and retraining logic for retraining the model using the training data comprising the image similarity training data and the social activity training data and the user feedback, the retraining logic comprising identifying logic for identifying, using the user feedback, a weighting for at least the image similarity training data and the social activity training data to be used in retraining the model.

13 . The system of claim 12 , the training data further comprising item metadata.

14 . The system of claim 8 , the generating logic for generating a set of recommended items using the trained model further comprising:
determining logic for determining, using the trained model, a click prediction for each of the items of the plurality, each item's click prediction is based at least on the item's features, the item's features comprising at least visual similarity and social activity features; and selecting logic for selecting one or more items of the plurality based on each item's click prediction.

15 . A computer readable non-transitory storage medium for tangibly storing thereon computer readable instructions that when executed cause at least one processor to:
determine visual similarity training data for a plurality of items, the visual similarity training data identifying, for each pair of items of the plurality of items, a level of visual similarity between images of the items in the pair; determine social activity training data for the plurality of items, the social activity training data identifying, for each pair of items of the plurality of items, an indicator of whether a shared user interest in the pair of items exists; train a model using a training data comprising the image similarity training data and the social activity training data; and generate a set of recommended items using the trained model.

16 . The computer readable non-transitory storage medium of claim 15 , the instructions to determine social activity training data further comprising instructions to determine the social training data using observed user behavior comprising one or more of a co-view, co-purchase and co-favorite observed user behavior.

17 . The computer readable non-transitory storage medium of claim 15 , the instructions to determine the image similarity training data further comprising instructions

Assignees

• Yahoo Inc

Inventors

• Hsiao Jen-Hao
• Liu Nan
• Li Jia

Classifications

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06K9/6256

  Physics · mapped topic