Machine optimization device, method and system
Machine optimization device, method and system
 CN 102,317,962 B
 Filed: 12/11/2009
 Issued: 03/16/2016
 Est. Priority Date: 12/12/2008
 Status: Active Grant
First Claim
1. the Computerized method for advertizer is mated with search terms, described method comprises:
 There is provided the first bigraph (bipartite graph) data structure, described first bigraph (bipartite graph) data structure has multiple advertizer'"'"'s node and multiple search terms node, and wherein each advertizer'"'"'s node is by the search terms node of edge conjunction to correspondence;
There is provided profit matrix, described profit matrix has the profit at the every bar edge for described first bigraph (bipartite graph) data structure;
There is provided the degree distribution data of the degree distribution representing each node, wherein, the degree distribution of each node represents the linking number preference of this node;
Produce the second bigraph (bipartite graph) data structure, described second bigraph (bipartite graph) data structure comprises the node of described first bigraph (bipartite graph) data structure and multiple dummy node;
The processor being suitable for performing weight matrixexpand is utilized to produce weight matrix, the value that described weight matrix comprises described profit matrix and the bonus values associated with described dummy node, the described bonus values in described weight matrix be determine based on described degree distribution data and comprise one group of weight of zero value;
The joint constraint of described first bigraph (bipartite graph) data structure will be corresponded in described second bigraph (bipartite graph) data structure to predetermined extent value, and do not retrain the described dummy node in described second bigraph (bipartite graph) data structure;
Mate based on described second bigraph (bipartite graph) data structure determination weight limit b, wherein b is arranged to described predetermined extent value;
Produce the middle bigraph (bipartite graph) data structure with binary weighted value;
To bear results graphic data structure to remove dummy node by intercepting described middle bigraph (bipartite graph) data structure;
Select predetermined quantity with advertizer'"'"'s node of each search terms node matching of one group of search terms node interested, described coupling determines based on the result pattern values of advertizer'"'"'s node adjacent with each interested search terms node;
AndExport the selected advertizer'"'"'s node with each interested search terms node matching.
Chinese PRB Reexamination
Abstract
Disclose method, system, computer program and computerreadable medium that a kind of producing level distributed intelligence is carried out mating.One embodiment of the method can comprise the graphic data structure execution b coupling to producing level distributed intelligence expansion, to identify the neighbours of selected input node.Belief propagation can be used to perform b coupling.Belief propagation method is suitable for using compressed message update rule and being suitable for use in distributed processing system(DPS).One embodiment can also comprise, by strengthening matching result to the first matching result level of application distributed intelligence to produce the second matching result.Disclose the embodiment for online advertisement/search terms coupling, Products Show, appointment service and social networks coupling, auction buyer/seller coupling and Resourse Distribute etc.
15 Claims

1. the Computerized method for advertizer is mated with search terms, described method comprises:

There is provided the first bigraph (bipartite graph) data structure, described first bigraph (bipartite graph) data structure has multiple advertizer'"'"'s node and multiple search terms node, and wherein each advertizer'"'"'s node is by the search terms node of edge conjunction to correspondence; There is provided profit matrix, described profit matrix has the profit at the every bar edge for described first bigraph (bipartite graph) data structure; There is provided the degree distribution data of the degree distribution representing each node, wherein, the degree distribution of each node represents the linking number preference of this node; Produce the second bigraph (bipartite graph) data structure, described second bigraph (bipartite graph) data structure comprises the node of described first bigraph (bipartite graph) data structure and multiple dummy node; The processor being suitable for performing weight matrixexpand is utilized to produce weight matrix, the value that described weight matrix comprises described profit matrix and the bonus values associated with described dummy node, the described bonus values in described weight matrix be determine based on described degree distribution data and comprise one group of weight of zero value; The joint constraint of described first bigraph (bipartite graph) data structure will be corresponded in described second bigraph (bipartite graph) data structure to predetermined extent value, and do not retrain the described dummy node in described second bigraph (bipartite graph) data structure; Mate based on described second bigraph (bipartite graph) data structure determination weight limit b, wherein b is arranged to described predetermined extent value; Produce the middle bigraph (bipartite graph) data structure with binary weighted value; To bear results graphic data structure to remove dummy node by intercepting described middle bigraph (bipartite graph) data structure; Select predetermined quantity with advertizer'"'"'s node of each search terms node matching of one group of search terms node interested, described coupling determines based on the result pattern values of advertizer'"'"'s node adjacent with each interested search terms node;
AndExport the selected advertizer'"'"'s node with each interested search terms node matching.


2. Computerized method according to claim 1, wherein determine that described weight limit b coupling comprises:

Utilize the processor passalong message between adjacent node being suitable for performing belief propagation vague generalization coupling, until meet termination condition, thus upgrade the trust value corresponding with each advertizer'"'"'s node being connected to selected search terms node, every bar message all based on the message of weight matrix value and reception, wherein determines the data content of every bar message according to compressed message update rule;
AndThe message that the trust value of each renewal and every bar receive is stored in the electronic memory associated with corresponding advertizer'"'"'s node.


3. method according to claim 2, is also included in the search results pages corresponding to the search terms node associated with selected advertizer'"'"'s node and shows each advertisement associated with selected advertizer'"'"'s node.

4. method according to claim 2, also be included in each advertizer'"'"'s node and each search terms Nodes stores a part for described profit matrix and a part for described degree distribution data, the adjacent node wherein based on each respective advertisement person'"'"'s node and each corresponding search terms node selects each part.

5. method according to claim 2, wherein utilizes parallel processing to perform described renewal and storage.

6. method according to claim 2, wherein said processor is the cloud computing system being suitable for using the belief propagation with degree distribution data to perform two couplings.

7. method according to claim 2, wherein said electronic memory is cloud storage system.

8. method according to claim 1, wherein each profit represents the profit with advertizer'"'"'s advertisementprinting of Advertisement association.

9. method according to claim 2, wherein performs described renewal repeatedly, until meet described termination condition.

10. method according to claim 2, wherein said termination condition is the repetition of the predetermined quantity of described renewal.

11. methods according to claim 2, wherein said termination condition is defined as the steady state (SS) of the trust value upgraded during predetermined a period of time.

12. methods according to claim 2, wherein said termination condition is the some message sent from each node.

13. methods according to claim 2, wherein said termination condition is predetermined a period of time in past.

14. methods according to claim 2, wherein said termination condition is defined as the steady state (SS) of the trust value upgraded during first predetermined a period of time and of occurring the earliest in the past second predetermined a period of time.

15. 1 kinds for the system of method according to claim 1 by advertisement and phrase match.
Specification(s)