Analyzing concepts over time

Patent metadata

Field	Value
Publication number	US-10671683-B2
Application number	US-201816119712-A
Country	US
Kind code	B2
Filing date	Aug 31, 2018
Priority date	Sep 22, 2015
Publication date	Jun 2, 2020
Grant date	Jun 2, 2020

Abstract

A method and apparatus are provided for automatically generating and processing first and second concept vector sets extracted, respectively, from a first set of concept sequences and from a second, temporally separated, concept sequences by performing a natural language processing (NLP) analysis of the first concept vector set and second concept vector set to detect changes in the corpus over time by identifying changes for one or more concepts included in the first and/or second set of concept sequences.

First claim

What is claimed is:

1. A method, in an information handling system comprising a processor and a memory, for analyzing concept vectors over time to detect changes in a corpus, the method comprising:
using a neutral network-based vector embedding method, matrix-based vector embedding method, log-linear classifier-based-based vector embedding method or word2vec method to generate a first concept vector set V′1, . . . , V′k+b derived from a concatenation of a first set of concept sequences and a second set of concept sequences over k concepts that are shared by the first and second sets of concept sequences and b concepts that are only in the second set of concept sequences, where the second set of concept sequences is effectively collected after collection of the first set of concept sequences;
detecting one or more disruptive concepts in the second set of concept sequences by analyzing relationship strengths between concepts to identify market trends for answering questions submitted to the information handling system, wherein analyzing relationship strengths comprises (a) computing cosine distances between (1) each concept vector V′k+1, . . . , V′k+b from the first concept vector set and (2) each of a second concept vector set VL1, . . . , VLh, derived from a third set of concept sequences identified in the concatenation of the first and second sets of concept sequences as being central to a specified technology area T over h concepts and (b) sorting vectors V′k+1, . . . , V′k+b to identify one or more disruptive concepts based on the computed cosine distances.

2. The method of claim 1 , wherein sorting vectors V′k+1, . . . , V′k+b comprises sorting the vectors V′k+1, . . . , V′k+b by a highest cosine distance to any of VL1, . . . , VLh in the second concept vector set.

3. The method of claim 2 , further comprising generating a list of R concepts corresponding to R sorted vectors V′k+1, . . . , V′k+b.

4. The method of claim 3 , further comprising reporting the R concepts as disruptive concepts.

5. The method of claim 1 , further comprising sorting vectors V′k+1, . . . , V′k+b from the first concept vector set to identify one or more additional disruptive concepts based on a computed cosine distance from high to low to a normalized sum of vectors (VL1+ . . . +VLh) from the second concept vector set.

6. The method of claim 5 , further comprising generating a list of R1 new concepts corresponding to R1 sorted vectors V′k+1, . . . , V′k+b.

7. The method of claim 6 , further comprising reporting the R1 new concepts as disruptive concepts.

8. The method of claim 1 , wherein analyzing relationship strengths comprises natural language processing (NLP) analysis to detect an appearance of one or more disruptive concepts in the second set of concept sequences that are related to a specified technology area represented by a sum of a plurality of concept vectors from the second concept vector set.

9. The method of claim 1 , wherein analyzing relationship strengths comprises performing natural language processing (NLP) analysis to detect an appearance of one or more disruptive concepts that have newly appeared with strong relationships to concepts in the third set of concept sequences that are central to the specified technology area T.

10. The method of claim 1 , wherein analyzing relationship strengths comprises performing natural language processing (NLP) analysis to detect differences in spatial and/or frequency distributions of first and second concept vector sets by identifying changes in values of quantitative geometry and topology features that characterize concept regions associated, respectively, with the first and second concept vector sets.

11. The method of claim 10 , wherein identifying changes comprises computing differences in centroid positions, diameters between extreme points, orientations of the principal axes, number of significant dimensions, aspect ratios between lengths of the principal axes, or number and sizes of clusters computed by standard clustering algorithms.

12. An information handling system comprising:
one or more processors;
a memory coupled to at least one of the processors;
a set of instructions stored in the memory and executed by at least one of the processors to analyze concept vectors over time to detect changes in a corpus, wherein the set of instructions are executable to perform actions of:
using a neutral network-based vector embedding method, matrix-based vector embedding method, log-linear classifier-based-based vector embedding method or word2vec method to generate a first concept vector set V′1, . . . , V′k+b derived from a concatenation of a first set of concept sequences and a second set of concept sequences over k concepts that are shared by the first and second sets of concept sequences and b concepts that are only in the second set of concept sequences, where the second set of concept sequences is effectively collected after collection of the first set of concept sequences;
detecting, by the system, one or more disruptive concepts in the second set of concept sequences by analyzing relationship strengths between concepts to identify market trends for answering questions submitted to the information handling system, wherein analyzing relationship strengths comprises (a) computing cosine distances between each concept vector V′k+1, . . . , V′k+b of the first concept vector set and each of a second vector set VL1, . . . , VLh derived from a third set of concept sequences identified in the concatenation of the first and second sets of concept sequences as being central to a specified technology area T over h concepts, and (b) sorting vectors V′k+1, . . . , V′k+b to identify one or more disruptive concepts based on the computed cosine distances.

13. The information handling system of claim 12 , wherein the set of instructions are executable to sort vectors V′k+1, . . . , V′k+b by sorting the vectors V′k+1, . . . , V′k+b by a highest cosine distance to any of VL1, . . . , VLh in the second concept vector set.

14. The information handling system of claim 13 , wherein the set of instructions are executable to generate a list of R concepts corresponding to R sorted vectors V′k+1, . . . , V′k+b.

15. The information handling system of claim 14 , wherein the set of instructions are executable to report the R concepts as disruptive concepts.

16. The information handling system of claim 12 , wherein the set of instructions are executable to sort vectors V′k+1, . . . , V′k+b from the first concept vector set to identify one or more additional disruptive concepts based on a computed cosine distance from high to low to a normalized sum of vectors (VL1+ . . . +VLh) from the second concept vector set.

17. The information handling system of claim 16 , wherein the set of instructions are executable to generate a list of R1 new concepts corresponding to R1 sorted vectors V′k+1, . . . , V′k+b.

18. The information handling system of claim 17 , wherein the set of instructions are executable to report the R1 new concepts as disruptive concepts.

19. A computer program product stored in a non-transitory computer readable storage medium, comprising computer instructions that, when executed by an information handling system, causes the system to analyze concept vectors over time to detect changes in a corpus by performing actions comprising:
using a neutral network-based vector embedding method, matrix-based vector embedding method, log-linear classifier-based-based vector embedding method or word2vec method to generate a first concept vec

Assignees

• IBM

Inventors

• Ho Tin Kam
• Lastras-Montano Luis A
• Shmueli Oded

Classifications

• G06F16/3344

  using natural language analysis · CPC title

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title

• G06F16/243

  Natural language query formulation · CPC title

• G06N3/08

  Learning methods · CPC title

• H05K999/99

  Electricity · mapped topic