Joint embedding of corpus pairs for domain mapping

• US 10,657,189 B2
• Filed: 08/18/2016
• Issued: 05/19/2020
• Est. Priority Date: 08/18/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Jagmohan, Ashish, Khabiri, Elham, Segal, Richard B., Vaculin, Roman
• Primary Examiner(s): Phan, Tuankhanh D
• Application Number:
  US15/240,660
• Publication Number:
  US 20180052924A1
• Time in Patent Office: 1,370 Days
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent

Abstract

Techniques for outside-in mapping for corpus pairs are provided. In one example, a computer-implemented method comprises: inputting first keywords associated with a first domain corpus; extracting a first keyword of the first keywords; inputting second keywords associated with a second domain corpus; generating an embedded representation of the first keyword via a trained model and generating an embedded representation of the second keywords via the trained model; and scoring a joint embedding affinity associated with a joint embedding. The scoring the joint embedding affinity comprises: transforming the embedded representation of the first keyword and the embedded representation of the second keywords via the trained model; determining an affinity value based on comparing the first keyword to the second keywords; and based on the affinity value, aggregating the joint embedding of the embedded representation of the first keyword and the embedded representation of the second keywords within the second domain corpus.

Claims

1. A system, comprising:

a memory that stores computer executable components; and

a processor that executes the computer executable components stored in the memory, wherein the computer executable components comprise;

a first learning component that;

analyzes first domain data associated with a domain comprising a first corpus, resulting in first analyzed data;

a second learning component that;

analyzes second domain data associated with a second domain comprising a second corpus, resulting in second analyzed data;

an identification component that;

based on the first analyzed data and the second analyzed data, identifies equivalent terms between the first domain data and the second domain data; and

a joint embedding component that;

based on the equivalent terms, the first analyzed data, and the second analyzed data, jointly embeds the first domain data and the second domain data, resulting in jointly embedded data and wherein the jointly embedding comprises analyzing a defined word, a sequence of words in a fixed window on either side of the defined word and generating a prediction of a context in which the defined word is likely to appear; and

in response to the jointly embedded data, outputs a model associated with the jointly embedded data employed to leverage across other domains to generate additional terms.

2. The system of claim 1, wherein the first domain data comprises a plurality of first keywords associated with a user identity of a user profile.

3. The system of claim 2, wherein the plurality of the first keywords comprise a second keyword associated with a web page.

4. The system of claim 1, wherein the joint embedding component also generates an affinity value related to the jointly embedded data.

5. The system of claim 4, wherein the affinity value is related to a score based on the joint embedding of the jointly embedded data.

6. The system of claim 2, wherein the second domain data comprises a second plurality of second keywords associated with a plurality of identities of a respective plurality of user profiles.

7. A computer program product for generating training data, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable to:

analyze first domain data associated with a domain comprising a first corpus, resulting in first analyzed data;

analyze second domain data associated with a second domain comprising a second corpus, resulting in second analyzed data;

based on the first analyzed data and the second analyzed data, identify equivalent terms between the first domain data and the second domain data;

based on the equivalent terms, the first analyzed data, and the second analyzed data, jointly embed the first domain data and the second domain data, resulting in jointly embedded data first domain data and the second domain data; and

a joint embedding component that;

based on the equivalent terms, the first analyzed data, and the second analyzed data, jointly embeds the first domain data and the second domain data, resulting in jointly embedded data and wherein the jointly embedded data is generated based on analyzing a defined word, a sequence of words in a fixed window on either side of the defined word and generating a prediction of a context in which the defined word is likely to appear; and

in response to the jointly embedded data, outputs a model associated with the jointly embedded data employed to leverage across other domains to generate additional terms; and

in response to the jointly embedded data, output a model associated with the jointly embedded data.

8. The computer program product of claim 7, wherein the program instructions are further executable to:

identify non-equivalent terms between the first domain data and the second domain data.

9. The computer program product of claim 7, wherein the first analyzed data comprises webpage data associated with a webpage.

10. The computer program product of claim 7, wherein the program instructions are further executable to:

rank the equivalent terms in descending order based on a frequency associated with the equivalent terms, resulting in a ranking.

11. The computer program product of claim 10, wherein the ranking comprises taxonomy tag data associated with a taxonomy tag of the equivalent terms.

12. The computer program product of claim 10, wherein the program instructions are further executable to:

output the ranking to the first corpus.

13. The computer program product of claim 7, wherein the program instructions are further executable to:

generate an affinity value in response to identifying the equivalent terms between the first domain data and the second domain data.

14. The computer program product of claim 13, wherein the affinity value is generated in response to an unsupervised estimation.

15. The computer program product of claim 7, wherein the program instructions are further executable to:

perform a cluster analysis of the first corpus relative to the second corpus.

16. A computer-implemented method, comprising:

analyzing, by a device operatively coupled to a processor, first domain data associated with a domain comprising a first corpus, resulting in first analyzed data;

analyzing, by the device, second domain data associated with a second domain comprising a second corpus, resulting in second analyzed data;

based on the analyzing the first domain data and the analyzing the second domain data, identifying, by the device, equivalent terms between the first domain data and the second domain data;

based on the equivalent terms, the first analyzed data, and the second analyzed data, jointly embedding, by the device, the first domain data and the second domain data, resulting in jointly embedded data and wherein the jointly embedded data is generated based on analyzing a defined word, a sequence of words in a fixed window on either side of the defined word and generating a prediction of a context in which the defined word is likely to appear; and

in response to the jointly embedding, outputting, by the device, a model associated with the jointly embedded data.

17. The computer-implemented method of claim 16, further comprising:

in response to the jointly embedding, scoring, by the device, the jointly embedded data.

18. The computer-implemented method of claim 17, further comprising:

in response to the scoring, generating by the device, an affinity value associated with the first domain data and the second domain data.

19. The computer-implemented method of claim 17, wherein an affinity value is related to the scoring of the jointly embedding of the first domain data and the second domain data.

20. The computer-implemented method of claim 16, further comprising:

generating a keyword ranking of keywords related to the first domain data and the second domain data, wherein the keyword ranking comprises taxonomy tag data associated with a taxonomy tag.