Method for estimating user interests

• US 9,740,782 B2
• Filed: 02/02/2015
• Issued: 08/22/2017
• Est. Priority Date: 09/12/2014
• Status: Active Grant

• Current Assignee: YE Hub Armenia LLC
• Original Assignee: Yandex Europe AG (Yandex N.V.)
• Inventors: Krasnikov, Andrey Borisovich
• Primary Examiner(s): Singh, Gurkanwaljit
• Application Number:
  US15/102,755
• Publication Number:
  US 20160314478A1
• Time in Patent Office: 932 Days
• Field of Search: None
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06Q 30/0201 Market modelling; Market analysis; Collecting market data
• CPC Class Codes: G06Q 30/0251 Targeted advertisements
• CPC Class Codes: H04B 7/0617 for beam forming
• CPC Class Codes: H04B 7/0626 Channel coefficients, e.g. channel state information [CSI]
• CPC Class Codes: H04L 41/046 comprising network management agents or mobile agents therefor
• CPC Class Codes: H04W 72/563 of the wireless resources

Abstract

Computer-implemented method for estimating user interests, executable by a computing device in communication with an output device, comprising: determining a first input vector corresponding to a first user event and a second input vector corresponding to a second user event; mapping first input vector to a first output vector and second input vector to a second 5 output vector in a first multidimensional space using a first vector-mapping module; determining a third input vector based on first output vector and second output vector; mapping third input vector to a third output vector in a second multidimensional space using a second vector-mapping module; determining a message to be provided to a user based on an analysis of at least one of first output vector and third output vector; and causing output 10 device to provide message to user. Also non-transitory computer-readable medium storing program instructions for carrying out the method.

Claims

1. A computer-implemented method for estimating user interests, the method executable by a computing device in communication with an output device, the method comprising:

determining, by the computing device, a first input vector corresponding to a first user event and a second input vector corresponding to a separate second user event;

configuring a first vector-mapping module to map given input vectors to output vectors in a first multidimensional space such that a respective distance separating the output vectors from each other is correlated to a difference between a user event context associated with the first input vector having been mapped to its output vector and a user event context associated with the second input vector having been mapped to its output vector, the configuring the first vector-mapping module including configuring a first neural network and a second neural network by;

connecting the first neural network and the second neural network in a coupled Siamese neural network arrangement, and

training the first and second neural networks to minimize a cross-modal loss between the first neural network and the second neural network;

mapping, using the first vector-mapping module of the computing device, the first input vector to a first output vector in the first multidimensional space and the separate second input vector to a second output vector in the first multidimensional space,

the first vector-mapping module being enabled to map both the first output vector and the second output vector to the first multidimensional space even if the first user event and the second user event are of different types;

determining, by the computing device, a third input vector based at least in part on the first output vector and the separate second output vector;

configuring a second vector-mapping module to map input vectors to output vectors in a second multidimensional space such that a respective distance separating the output vectors from each other is correlated to a difference between a user event context associated with one input vector, the one input vector being a respective output vector of the first vector-mapping module and a user event context associated with another input, the other input vector being another respective output vector of the first vector-mapping module, the configuring the second vector-mapping module including configuring a third neural network;

mapping, using the second vector-mapping module of the computing device, the third input vector to a third output vector in the second multidimensional space;

determining, by the computing device, a message to be provided to a user based at least in part on an analysis of at least one of the first output vector and the third output vector; and

causing, by the computing device, the output device to provide the message to the user.

2. The method of claim 1, wherein:

the first vector-mapping module has been configured to map first configuration input vectors to first configuration output vectors in the first multidimensional space such that a respective distance separating each one of the first configuration output vectors from each other one of the first configuration output vectors is correlated to a difference between a context associated with a respective one of the first configuration input vectors having been mapped to the one of the first configuration output vectors and a context associated with a respective other one of the first configuration input vectors having been mapped to the other one of the first configuration output vectors; and

the second vector-mapping module has been configured to map second configuration input vectors to second configuration output vectors in the second multidimensional space such that a respective distance separating each one of the second configuration output vectors from each other one of the second configuration output vectors is correlated to a difference between a context associated with a respective one of the second configuration input vectors having been mapped to the one of the second configuration output vectors and a context associated with a respective other one of the second configuration input vectors having been mapped to the other one of the second configuration output vectors.

3. The method of claim 2, wherein:

the difference between the context associated with the respective one of the first configuration input vectors having been mapped to the one of the first configuration output vectors and the context associated with the respective other one of the first configuration input vectors having been mapped to the other one of the first configuration output vectors comprises a time interval between a time associated with the respective one of the first configuration input vectors having been mapped to the one of the first configuration output vectors and a time associated with the respective other one of the first configuration input vectors having been mapped to the other one of the first configuration output vectors; and

the difference between the context associated with the respective one of the second configuration input vectors having been mapped to the one of the second configuration output vectors and the context associated with the respective other one of the second configuration input vectors having been mapped to the other one of the second configuration output vectors comprises a time interval between a time associated with the respective one of the second configuration input vectors having been mapped to the one of the second configuration output vectors and a time associated with the respective other one of the second configuration input vectors having been mapped to the other one of the second configuration output vectors.

4. The method of claim 1, further comprising:

determining a fourth input vector based at least in part on the third output vector; and

mapping the fourth input vector to a fourth output vector in a third multidimensional space using a third vector-mapping module.

5. The method of claim 4, wherein:

the first vector-mapping module has been configured to map first configuration input vectors to first configuration output vectors in the first multidimensional space such that a respective distance separating each one of the first configuration output vectors from each other one of the first configuration output vectors is correlated to a difference between a context associated with a respective one of the first configuration input vectors having been mapped to the one of the first configuration output vectors and a context associated with a respective other one of the first configuration input vectors having been mapped to the other one of the first configuration output vectors;

the second vector-mapping module has been configured to map second configuration input vectors to second configuration output vectors in the second multidimensional space such that a respective distance separating each one of the second configuration output vectors from each other one of the second configuration output vectors is correlated to a difference between a context associated with a respective one of the second configuration input vectors having been mapped to the one of the second configuration output vectors and a context associated with a respective other one of the second configuration input vectors having been mapped to the other one of the second configuration output vectors; and

the third vector-mapping module has been configured to map third configuration input vectors to third configuration output vectors in the third multidimensional space such that a respective distance separating each one of the third configuration output vectors from each other one of the third configuration output vectors is correlated to a difference between a context associated with a respective one of the third configuration input vectors having been mapped to the one of the third configuration output vectors and a context associated with a respective other one of the third configuration input vectors having been mapped to the other one of the third configuration output vectors.

6. The method of claim 5, wherein:

the difference between the context associated with the respective one of the first configuration input vectors having been mapped to the one of the first configuration output vectors and the context associated with the respective other one of the first configuration input vectors having been mapped to the other one of the first configuration output vectors comprises a time interval between a time associated with the respective one of the first configuration input vectors having been mapped to the one of the first configuration output vectors and a time associated with the respective other one of the first configuration input vectors having been mapped to the other one of the first configuration output vectors;

the difference between the context associated with the respective one of the second configuration input vectors having been mapped to the one of the second configuration output vectors and the context associated with the respective other one of the second configuration input vectors having been mapped to the other one of the second configuration output vectors comprises a time interval between a time associated with the respective one of the second configuration input vectors having been mapped to the one of the second configuration output vectors and a time associated with the respective other one of the second configuration input vectors having been mapped to the other one of the second configuration output vectors; and

the difference between the context associated with the respective one of the third configuration input vectors having been mapped to the one of the third configuration output vectors and the context associated with the respective other one of the third configuration input vectors having been mapped to the other one of the third configuration output vectors comprises a time interval between a time associated with the respective one of the third configuration input vectors having been mapped to the one of the third configuration output vectors and a time associated with the respective other one of the third configuration input vectors having been mapped to the other one of the third configuration output vectors.

7. The method of claim 4, wherein:

the third output vector represents a plurality of user events having occurred during a first period of time; and

the fourth output vector represents a plurality of user events having occurred during a second period of time, the second period of time being longer than the first period of time and including the first period of time.

8. The method of claim 1, wherein the first user event is of a first user event type and the second user event is of a second user event type other than the first user event type;

and

wherein;

mapping the first input vector to the first output vector in the first multidimensional space using the first vector-mapping module comprises mapping the first input vector to the first output vector using a first neural network configured to map input vectors corresponding to user events of the first user event type to output vectors in the first multidimensional space;

mapping the second input vector to the second output vector in the first multidimensional space using the first vector-mapping module comprises mapping the second input vector to the second output vector using a second neural network configured to map input vectors corresponding to user events of the second user event type to output vectors in the first multidimensional space; and

mapping the third input vector to the third output vector in the second multidimensional space using the second vector-mapping module comprises mapping the third input vector to the third output vector in the second multidimensional space using a third neural network having been configured to map input vectors having been determined based on at least one vector in the first multidimensional space to output vectors in the second multidimensional space.

9. The method of claim 3, further comprising:

before mapping the first input vector to the first output vector using the first vector-mapping module, configuring the first vector-mapping module to map the first configuration input vectors to the first configuration output vectors in the first multidimensional space such that the respective distance separating each one of the first configuration output vectors from each other one of the first configuration output vectors is correlated to the time interval between the time associated with the respective one of the first configuration input vectors being mapped to the one of the first configuration output vectors and the time associated with the respective other one of the first configuration input vectors being mapped to the other one of the first configuration output vectors; and

before mapping the third input vector to the third output vector using the second vector-mapping module, configuring the second vector-mapping module to map the second configuration input vectors to the second configuration output vectors in the second multidimensional space such that the respective distance separating each one of the second configuration output vectors from each other one of the second configuration output vectors is correlated to the time interval between the time associated with the respective one of the second configuration input vectors being mapped to the one of the second configuration output vectors and the time associated with the respective other one of the second configuration input vectors being mapped to the other one of the second configuration output vectors.

10. The method of claim 1, wherein the first user event and the second user event are associated with the user.

11. The method of claim 1, wherein at least one of the first user event and the second user event is associated with a second user other than the user.

12. The method of claim 10, wherein the first configuration input vectors correspond to user events associated with the user.

13. The method of claim 11, wherein at least one of the first configuration input vectors corresponds to a user event associated with a second user other than the user.

14. The method of claim 1, further comprising receiving an indication of the first user event from a user input device in communication with the computing device.

15. The method of claim 14, further comprising receiving at least one of:

an indication of the second user event from the user input device; and

an indication of the second user event from a second user input device other than the user input device.

16. The method of claim 1, wherein the first user event comprises a use of a network resource.

17. The method of claim 16, wherein the network resource comprises a search engine and the use of the network resource comprises a submission of a search query.

18. The method of claim 17, wherein determining the first input vector corresponding to the first user event comprises determining the first input vector based at least in part on a search term of the search query.

19. The method of claim 1, wherein the second user event comprises an association of a user with a geographical location.

20. The method of claim 1, wherein:

determining the message to be provided to the user comprises estimating a network resource to be of interest to the user; and

causing the output device to provide the message to the user comprises causing the output device to provide an indication of the network resource to the user.

21. The method of claim 1, wherein determining the message to be provided to the user based on the analysis of the at least one of first output vector and the third output vector comprises:

feeding the at least one of the first output vector and the third output vector as an input to an interest identification module having been configured to identify user interests based on vectors in at least one of the first multidimensional space and the second multidimensional space; and

determining the message to be provided based on an interest of the user identified by the interest identification module.

22. The method of claim 1, wherein determining the message to be provided to the user based on the analysis of the at least one of first output vector and the third output vector comprises:

identifying a second user similar to the user based on the analysis;

identifying an interest of the second user; and

determining the message to be provided to the user based on the interest of the second user.

23. The method of claim 1, wherein determining the message to be provided to the user based on the analysis of the at least one of first output vector and the third output vector comprises:

associating the user with a group of users based on the analysis;

identifying an interest of the group of users; and

determining the message to be provided to the user based on the interest of the group of users.

24. The method of claim 1, wherein determining the third input vector based at least in part on the first output vector and the second output vector comprises calculating an average of vectors including the first output vector and the second output vector.

25. A non-transitory computer-readable medium storing program instructions for estimating user interests, the program instructions being executable by a computing device in communication with an output device to effect:

determination, by the computing device, of a first input vector corresponding to a first user event and a second input vector corresponding to a separate second user event;

configuring a first vector-mapping module to map given input vectors to output vectors in a first multidimensional space such that a respective distance separating the output vectors from each other is correlated to a difference between a user event context associated with the first input vector having been mapped to its output vector and a user event context associated with the second input vector having been mapped to its output vector, the configuring the first vector-mapping module including configuring a first neural network and a second neural network by;

connecting the first neural network and the second neural network in a coupled Siamese neural network arrangement, and

training the first and second neural networks to minimize a cross-modal loss between the first neural network and the second neural network;

mapping, using a first vector-mapping module of the computing device, of the first input vector to a first output vector in a first multidimensional space and the separate second input vector to a second output vector in the first multidimensional space,

the first vector-mapping module being enabled to map both the first output vector and the second output vector to the first multidimensional space even if the first user event and the second user event are of different types;

determination, by the computing device, of a third input vector based at least in part on the first output vector and the separate second output vector;

configuring a second vector-mapping module to map input vectors to output vectors in a second multidimensional space such that a respective distance separating the output vectors from each other is correlated to a difference between a user event context associated with one input vector, the one input vector being a respective output vector of the first vector-mapping module and a user event context associated with another input, the other input vector being another respective output vector of the first vector-mapping module, the configuring the second vector-mapping module including configuring a third neural network;

mapping, using the second vector-mapping module of the computing device, of the third input vector to a third output vector in the second multidimensional space;

determination, by the computing device, of a message to be provided to a user based at least in part on an analysis of at least one of the first output vector and the third output vector; and

causing, by the computing device, the output device to provide the message to the user.