Method, device and equipment for fusing different instances describing same entity

1. A computing device, comprising:

a memory for storing a connection diagram comprising a plurality of instances, wherein different nodes in the connection diagram represent different instances, and connection lines between the nodes represent instance relations between the instances corresponding to the nodes; and

a processor connected to the memory, wherein the processor is configured to acquire the connection diagram from the memory, and the processor is configured to:

based on the instance relations, identify different instances describing the same entity in the connection diagram, fuse the nodes corresponding to the identified instances, and update the connection diagram;

in the updated connection diagram, identify an instance pair with an unidentified instance relation, and add a connection line to connect the nodes corresponding to instances in the instance pair;

iteratively perform the step of updating the connection diagram based on the instance relations and the operation of adding a connection line in the updated connection diagram until a specified condition is satisfied.

2. A computer program product for fusing different instances describing the same entity, the computer program product comprising non-transitory computer readable medium storing computer program instructions for:

acquiring a connection diagram comprising a plurality of instances, wherein different nodes in the connection diagram represent different instances, and connection lines between the nodes represent instance relations between the instances corresponding to the nodes;

based on the instance relations, identifying different instances describing the same entity in the connection diagram, fusing the nodes corresponding to the identified instances, and updating the connection diagram;

in the updated connection diagram, identifying an instance pair with an unidentified instance relation, and adding a connection line to connect the nodes corresponding to instances in the instance pair; and

iteratively performing the operation of updating the connection diagram and the operation of adding a connection line until a specified condition is satisfied.

3. The computer program product according to claim 2 , wherein the non-transitory computer readable medium stores the computer program instructions for:

calculating a correlation between a first instance corresponding to a first node and a second instance corresponding to a second node connected to the first node through N nodes in the updated connection diagram, wherein N is greater than or equal to 1;

identifying an instance pair corresponding to two nodes whose correlation reaches a predetermined correlation threshold to be an instance pair having an instance relation, and adding a connection line to connect the two nodes.

4. The computer program product according to claim 2 , wherein the specified condition is set as:

the number of unidentified instance pairs with instance relation in the updated connection diagram is zero.

5. The computer program product according to claim 2 , wherein the non-transitory computer readable medium stores the computer program instructions for:

grouping the plurality of instances;

calculating a similarity between any two instances in a group based on instance relation thereof for each group; and

identifying an instance pair whose similarity reaches a predetermined similarity threshold to be an instance pair describing the same entity.

6. The computer program product according to claim 5 , wherein, for two instances from different sources, the non-transitory computer readable medium stores the computer program instructions for calculating a similarity Sim between the two instances according to the formula shown below:

Sim

=

Jac

ij

/

Uniq

Jac

ij

=

C

i

⋂

C

j

C

i

⋃

C

j

Uniq

=

Log

⁡

(

Max

⁡

(

Cnt

source

⁢

⁢

A

,

i

,

Cnt

source

⁢

⁢

B

,

j

)

+

1

)

wherein C i is an instance set has an insurance relation with an instance i, wherein C i is an instance set has an insurance relation with an instance j, wherein Jac ij is an instance relation similarity between i and j, wherein Uniq is a measure for uniqueness of an instance, wherein Cnt sourceA,i is the number of same name instances of instance i in the source A, and wherein Cnt sourceB,j is the number of same name instances of instance j in the source B.

7. The computer program product according to claim 2 , wherein the non-transitory computer readable medium stores the computer program instructions for:

calculating an attribute similarity between instances corresponding to any two nodes in the connection diagram; and

fusing nodes corresponding to two instances whose attribute similarity exceeds a predetermined attribute similarity threshold into one node.

8. A method for fusing different instances describing the same entity, comprising:

acquiring a connection diagram comprising a plurality of instances, wherein different nodes in the connection diagram represent different instances, and connection lines between the nodes represent instance relations between the instances corresponding to the nodes;

based on the instance relations, identifying different instances describing the same entity in the connection diagram, fusing the nodes corresponding to the identified instances, and updating the connection diagram;

in the updated connection diagram, identifying an instance pair with an unidentified instance relation, and adding a connection line to connect the nodes corresponding to instances in the instance pair; and

iteratively performing the step of updating the connection diagram based on the instance relation and the operation of adding a connection line in the updated connection diagram until a specified condition is satisfied.

9. The method according to claim 8 , wherein the step of, in the updated connection diagram, identifying an instance pair with an unidentified instance relation comprises:

calculating a correlation between a first instance corresponding to a first node and a second instance corresponding to a second node connected to the first node through N nodes in the updated connection diagram, wherein N is greater than or equal to 1; and

identifying an instance pair corresponding to two nodes whose correlation reaches a predetermined correlation threshold to be an instance pair having an instance relation, and adding a connection line to connect the two nodes.

10. The method according to claim 8 , wherein the specified condition is set as:

the number of unidentified instance pairs with instance relation in the updated connection diagram is zero.

11. The method according to claim 8 , wherein the step of, based on the instance relations, identifying different instances describing the same entity in the connection diagram comprises:

grouping the plurality of instances;

calculating a similarity between any two instances in a group based on instance relation thereof for each group; and

identifying an instance pair whose similarity reaches a predetermined similarity threshold to be an instance pair describing the same entity.

12. The method according to claim 11 , further comprising, for two instances from different sources, calculating a similarity Sim between the two instances according to the formula shown below:

Sim

=

Jac

ij

/

Uniq

Jac

ij

=

C

i

⋂

C

j

C

i

⋃

C

j

Uniq

=

Log

⁡

(

Max

⁡

(

Cnt

source

⁢

⁢

A

,

i

,

Cnt

source

⁢

⁢

B

,

j

)

+

1

)

wherein C i is an instance set has an insurance relation with an instance i, wherein C j is an instance set has an insurance relation with an instance j, wherein Jac ij is an instance relation similarity between i and j, Uniq is a measure for uniqueness of an instance, wherein Cnt sourceA,i is the number of same name instances of instance i in the source A, and wherein Cnt sourceB,j is the number of same name instances of instance j in the source B.

13. The method according to claim 8 , wherein the step of acquiring the connection diagram comprising a plurality of instances comprises:

calculating an attribute similarity between instances corresponding to any two nodes in the connection diagram; and

fusing nodes corresponding to two instances whose attribute similarity exceeds a predetermined attribute similarity threshold into one node.