System and method for incorporating new terms in a term-vector space from a semantic lexicon

• US 20170161275A1
• Filed: 12/08/2015
• Published: 06/08/2017
• Est. Priority Date: 12/08/2015
• Status: Abandoned Application

• Current Assignee: Luminoso Technologies, Inc.
• Original Assignee: Luminoso Technologies, Inc.
• Inventors: Speer, Robert
• Application Number:
  US14/962,007
• Publication Number:
  US 20170161275A1
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/2282 Tablespace storage structures; Management thereof
• CPC Class Codes: G06F 16/243 Natural language query formulation
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries
• CPC Class Codes: G06F 16/319 Inverted lists
• CPC Class Codes: G06F 16/328 Management therefor

Abstract

A method for incorporating new terms in a term-vector space from a semantic lexicon includes identifying, by a computing device, a first term, the first term present in a first semantic lexicon, the first term absent from a term vector space. The method includes obtaining, by the computing device, from the first semantic lexicon, at least one second term related to the first term in the first semantic lexicon. The method includes finding, by the computing device, at least one vector in the vector space corresponding to the at least one second term. The method includes generating, by the computing device, a vector corresponding to the first term using the at least one vector corresponding to the at least one second term.

Claims

1. A method for incorporating new terms in a term vector space from a semantic lexicon, the method comprising:

identifying, by a computing device, a first term, the first term present in a first semantic lexicon, the first term absent from a term vector space represented by a term vector matrix;

obtaining, by the computing device, from the first semantic lexicon, at least one second term related to the first term in the first semantic lexicon;

finding, by the computing device, at least one vector in the vector space corresponding to the at least one second term; and

generating, by the computing device, a vector corresponding to the first term using the at least one vector corresponding to the at least one second term.

2. The method of claim 1, wherein identifying further comprises determining that the first term has more than a threshold number of connections to other terms within the first semantic lexicon

3. The method of claim 1, wherein obtaining further comprises determining that the at least one second term and the first term have a connection weight exceeding a threshold number.

4. The method of claim 1, wherein the at least one second term is a plurality of second terms, wherein the at least one second vector is a plurality of second vectors, each second vector corresponding to a term of the plurality of second terms, and wherein generating further comprises combining the plurality of second vectors together to generate the vector corresponding to the first term.

5. The method of claim 4, wherein combining the plurality of second vectors further comprises computing a mean of the plurality of second vectors.

6. The method of claim 5, wherein computing the mean further comprises:

calculating a degree of similarity between the first term and each second term; and

weighting each second vector of the plurality of second vectors by the degree of similarity between the first term and the second term corresponding to the second vector.

7. The method of claim 6, wherein calculating the degree of similarity further comprises obtaining a relatedness confidence score.

8. The method of claim 4, wherein combining the plurality of second vectors further comprises weighting each second vector of the plurality of second vectors by a reliability score.

9. The method of claim 1 further comprising performing column normalization of the term vector matrix.

10. The method of claim 1 further comprising performing row normalization of the term vector matrix.

11. The method of claim 1, further comprising retrofitting the term vector space to the first semantic lexicon, producing a retrofitted term vector matrix.

12. The method of claim 11, wherein retrofitting further comprises computing a product of the term vector space with a matrix representing the first semantic lexicon.

13. The method of claim 12 further comprising adding the retrofitted term vector matrix to the term vector matrix to produce an intermediate matrix, and computing the product of the intermediate matrix with the matrix representing the first semantic lexicon.

14. The method of claim 12, wherein the matrix representing the first semantic lexicon is a square matrix having a plurality of diagonal cells, and further comprising weighting each diagonal cell of the plurality of diagonal cells.

15. The method of claim 11 further comprising retrofitting the retrofitted term vector matrix to the first semantic lexicon.

16. The method of claim 11 further comprising computing the mean of each vector in the term vector space with itself, and replacing each vector in the term vector space with the computed mean.

17. The method of claim 11 further comprising retrofitting the retrofitted term vector space to a second semantic lexicon.

18. The method of claim 1, further comprising:

identifying a plurality of terms in the term vector space that correspond to a single term in the first semantic lexicon; and

combining a plurality of vectors representing the plurality of terms together into a single vector representing the single term.

19. The method of claim 18, wherein combining further comprises computing a weighted average of the plurality of vectors.

20. The method of claim 1 further comprising generating the first semantic lexicon by combining a second semantic lexicon and a third semantic lexicon.

21. A system for incorporating new terms in a term-vector space from a lexicon, the system comprising:

a term vector space;

a first semantic lexicon; and

a computing device, the computing device configured to identify a first term, the first term present in the first semantic lexicon, the first term absent from the term vector space, to obtain from the first semantic lexicon, at least one second term related to the first term in the first semantic lexicon, to find at least one vector in the vector space corresponding to the at least one second term, and to generate a vector corresponding to the first term using the at least one vector corresponding to the at least one second term.