Analysis apparatus, analysis method and program

1. An analysis device comprising:

a vector estimator configured to estimate, in response to receiving input of a plurality of pieces of relational data each including a plurality of objects and a relationship between the objects, for each piece of relational data, a latent vector for characterizing a structure of the relational data by using the objects and the relationship included in the relational data; and

a matcher configured to match, for each set of first relational data and second relational data different from each other of the received pieces of relational data, a first object and a second object by using a first latent vector corresponding to the first object included in the first relational data and a second latent vector corresponding to the second object included in the second relational data, wherein the vector estimator estimates, when the relationship indicates a close relationship between a plurality of objects to each other, latent vectors corresponding to the objects such that the latent vectors corresponding to the respective objects have a close relationship to each other.

2. The analysis device according to claim 1 , comprising:

a transformation matrix estimator configured to estimate, for each piece of relational data, a transformation matrix for transforming the latent vector such that distributions of the latent vectors are close to each other between the received pieces of relational data; and

a transformer configured to transform, for each piece of relational data, a latent vector corresponding to an object included in the piece of relational data into a transformed latent vector by a transformation matrix corresponding to the piece of relational data, wherein the matcher matches, for each set of first relational data and second relational data different from each other of the received pieces of relational data, the first object and the second object by using a first transformed latent vector corresponding to the first object and a second transformed latent vector corresponding to the second object.

3. The analysis device according to claim 2 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

4. The analysis device according to claim 2 , wherein the distributions of the latent vectors are closer to each other by reducing an maximum mean discrepancy (MMD) based on a distance between the distributions.

5. The analysis device according to claim 1 , wherein the vector estimator estimates, when the relationship indicates a close relationship between a plurality of objects to each other, latent vectors corresponding to the objects such that the latent vectors corresponding to the respective objects have a close relationship to each other and distributions of the latent vectors are close to each other between the received pieces of relational data.

6. The analysis device according to claim 5 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

7. The analysis device according to claim 1 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

8. An analysis method executed by a computer, the method comprising:

estimating, by a vector estimator, in response to receiving input of a plurality of pieces of relational data each including a plurality of objects and a relationship between the objects, for each piece of relational data, a latent vector for characterizing a structure of the relational data by using the objects and the relationship included in the relational data; and

matching, by a matcher, for each set of first relational data and second relational data different from each other of the received pieces of relational data, a first object and a second object by using a first latent vector corresponding to the first object included in the first relational data and a second latent vector corresponding to the second object included in the second relational data, wherein the vector estimation step includes estimating, when the relationship indicates a close relationship between a plurality of objects to each other, latent vectors corresponding to the objects such that the latent vectors corresponding to the respective objects have a close relationship to each other.

9. The analysis method according to claim 8 , the method further comprising:

estimating, by a transformation matrix estimator, for each piece of relational data, a transformation matrix for transforming the latent vector such that distributions of the latent vectors are close to each other between the received pieces of relational data; and

transforming, by a transformer, for each piece of relational data, a latent vector corresponding to an object included in the piece of relational data into a transformed latent vector by a transformation matrix corresponding to the piece of relational data, wherein the matcher matches, for each set of first relational data and second relational data different from each other of the received pieces of relational data, the first object and the second object by using a first transformed latent vector corresponding to the first object and a second transformed latent vector corresponding to the second object.

10. The analysis method according to claim 9 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

11. The analysis method according to claim 9 , wherein the distributions of the latent vectors are closer to each other by reducing an maximum mean discrepancy (MMD) based on a distance between the distributions.

12. The analysis method according to claim 8 , wherein the vector estimator estimates, when the relationship indicates a close relationship between a plurality of objects to each other, latent vectors corresponding to the objects such that the latent vectors corresponding to the respective objects have a close relationship to each other and distributions of the latent vectors are close to each other between the received pieces of relational data.

13. The analysis method according to claim 12 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

14. The analysis method according to claim 8 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

15. A system for analyzing, the system comprises:

a processor; and

a memory storing computer-executable program instructions that when executed by the processor cause the system to:

estimate, by a vector estimator, in response to receiving input of a plurality of pieces of relational data each including a plurality of objects and a relationship between the objects, for each piece of relational data, a latent vector for characterizing a structure of the relational data by using the objects and the relationship included in the relational data; and

match, by a matcher, for each set of first relational data and second relational data different from each other of the received pieces of relational data, a first object and a second object by using a first latent vector corresponding to the first object included in the first relational data and a second latent vector corresponding to the second object included in the second relational data, wherein the vector estimation step includes estimating, when the relationship indicates a close relationship between a plurality of objects to each other, latent vectors corresponding to the objects such that the latent vectors corresponding to the respective objects have a close relationship to each other.

16. The system for analyzing according to claim 15 , the computer-executable program instructions when executed further causing the system to:

estimate, by a transformation matrix estimator, for each piece of relational data, a transformation matrix for transforming the latent vector such that distributions of the latent vectors are close to each other between the received pieces of relational data; and

transform, by a transformer, for each piece of relational data, a latent vector corresponding to an object included in the piece of relational data into a transformed latent vector by a transformation matrix corresponding to the piece of relational data, wherein the matcher matches, for each set of first relational data and second relational data different from each other of the received pieces of relational data, the first object and the second object by using a first transformed latent vector corresponding to the first object and a second transformed latent vector corresponding to the second object.

17. The system for analyzing according to claim 16 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

18. The system for analyzing according to claim 15 , wherein the vector estimator_estimates, when the relationship indicates a close relationship between a plurality of objects to each other, latent vectors corresponding to the objects such that the latent vectors corresponding to the respective objects have a close relationship to each other and distributions of the latent vectors are close to each other between the received pieces of relational data.

19. The system for analyzing according to claim 18 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.

20. The system for analyzing according to claim 15 , wherein the vector estimator estimates the latent vector by a neural network that has been learned in advance.