Extraction of knowledge points and relations from learning materials

• US 9,852,648 B2
• Filed: 07/10/2015
• Issued: 12/26/2017
• Est. Priority Date: 07/10/2015
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Wang, Jun, Uchino, Kanji
• Primary Examiner(s): Yao, Sam
• Assistant Examiner(s): Humphrey, Michael
• Application Number:
  US14/796,838
• Publication Number:
  US 20170011642A1
• Time in Patent Office: 900 Days
• Field of Search: 434322
• CPC Class Codes: G06F 16/316 Indexing structures
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06F 16/367 Ontology
• CPC Class Codes: G06F 16/95 Retrieval from the web
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G09B 5/00 Electrically-operated educational appliances working with questions and answers G09B7/00; simulators G09B9/00; advertising or displaying in general G09F
• CPC Class Codes: G09B 7/00 Electrically-operated teaching apparatus or devices working with questions and answers mechanically operated G09B3/00; computing arrangements G06F

Abstract

A method of automated domain knowledge structure generation includes crawling learning materials. The method may include extracting structural information from the learning materials. The method may include extracting knowledge points from the learning materials. The method may include inferring dependency relationships between the knowledge points. The method may include aligning one or more of the knowledge points with one or more of the learning materials. The method may also include generating a domain knowledge structure. The domain knowledge structure may include the extracted knowledge points organized at least partially according to the inferred hierarchy and dependency relationships. The extracted knowledge points may include the aligned learning materials.

Claims

1. A method of domain knowledge structure generation in online learning materials, the method comprising:

crawling, by one or more processors, electronic learning materials stored at least temporarily in one or more non-transitory storage media;

extracting, by the one or more processors, structural information from the electronic learning materials;

extracting, by the one or more processors, knowledge points from the electronic learning materials based at least partially on an organization of the electronic learning materials as indicated by the extracted structural information;

locating, by the one or more processors, positions of each of the knowledge points within each of the learning materials;

comparing, by the one or more processors, relative positions of knowledge points in the learning materials in which the knowledge points are present;

inferring, by the one or more processors, hierarchy and dependency relationships between the knowledge points based at least partially on the relative positions of the knowledge points;

aligning, by the one or more processors, one or more of the knowledge points with at least one specific portion of one or more of the learning materials, wherein each of the specific portions include one of the located positions of the corresponding knowledge point within the learning materials; and

generating, by the one or more processors, a domain knowledge structure including the extracted knowledge points organized at least partially according to the inferred hierarchy and dependency relationships and including the aligned learning materials.

2. The method of claim 1, wherein the learning materials include one or more of open courseware, massive open courses (MOOC), personal home pages, department home pages, and electronic books (e-books).

3. The method of claim 1, wherein the structural information are extracted from learning materials metadata of the electronic learning materials and semi-structured format information included in the electronic learning materials.

4. The method of claim 1, wherein the extracting structural information includes:

extracting lists from one or more of a syllabus, lecture notes, and a table of contents included in the electronic learning materials;

conducting a page-format analysis of one or more pages of the electronic learning materials;

detecting sequence borders based on one or more of line breaks, table cell borders, sentence borders, and specific punctuation; and

generating segmented term sequences from the lists and the page-format analysis, wherein one or more of the segmented term sequences are bounded by the detected sequence borders include positional information indicating a position in the electronic learning materials of the segmented term sequences.

5. The method of claim 4, wherein the extracting knowledge points includes:

receiving the segmented term sequences having positional information indicating a position in the electronic learning materials of the segmented term sequence;

after the receiving the segmented term sequences, unifying abbreviations of the segmented term sequences;

after the unifying the abbreviations, constructing generalized suffix trees of the segmented term sequences;

after the constructing the generalized suffix trees, discovering repeated phrase instances of the segmented term sequences using the generalized suffix trees, wherein the phrase instances are limited by a particular maximum length;

after the discovering the repeated phrase instances, adjusting frequency of the discovered repeated phrases instances based on positions of phrase instances in the learning materials;

after the adjusting the frequency, measuring a cohesion and a separation of the segmented term sequences;

after the measuring the cohesion and the separation, removing stop phrases from the segmented term sequences;

after the removing the stop phrases, generating candidate knowledge points from the segmented term sequences;

after the generating the candidate knowledge points, calculating weights of the candidate knowledge points based on the adjusted frequency;

after the calculating the weights, analyzing appearance positions of the candidate knowledge points;

after the analyzing the appearance positions, constructing a hierarchy of knowledge points based on the appearance positions; and

after the constructing the hierarchy, presenting coverage overview of the learning materials.

6. The method of claim 5, wherein:

the cohesion is measured according to a mutual information cohesion metric;

the separation is measured according to an accessor variety separation metric;

the abbreviations are unified according to a principal component analysis or a singular value decomposition; and

the weights are calculated according to a number of appearances and an authority of a particular learning material in which the segmented term sequences appear.

7. The method of claim 5, wherein a hierarchy level of each of the knowledge points is based on a level of granularity of the knowledge points in the learning materials.

8. The method of claim 5, wherein the inferring dependency includes:

deciding the relative positions between two knowledge points in the learning materials;

recommending immediate dependency relationships based on distance between the knowledge points in the hierarchy of knowledge points; and

generating a knowledge point map representative of the dependency relationships.

9. The method of claim 1, wherein the aligning one or more of the knowledge points includes:

precisely locating each of the knowledge points in the learning materials based on granularity of the knowledge point; and

comparing and ranking the learning materials aligned with specific knowledge points.

10. The method of claim 9, wherein the learning materials are compared and ranked according to a linear combination of a first weight multiplied by a knowledge point score, a second weight multiplied by a general score, and a third weight multiplied by a type factor.

11. A non-transitory computer-readable medium having encoded therein programming code executable by one or more processors to perform or control performance of operations comprising:

crawling electronic learning materials stored at least temporarily in one or more non-transitory storage media;

extracting structural information from the electronic learning materials;

extracting knowledge points from the electronic learning materials based at least partially on an organization of the electronic learning materials as indicated by the extracted structural information;

locating positions of each of the knowledge points within each of the learning materials;

comparing relative positions of knowledge points in the learning materials in which the knowledge points are present;

inferring hierarchy and dependency relationships between the knowledge points based at least partially on the relative positions of the knowledge points;

aligning one or more of the knowledge points with at least one specific portion of one or more of the learning materials, wherein each of the specific portions include one of the located positions of the corresponding knowledge point within the learning materials; and

generating a domain knowledge structure including the extracted knowledge points organized at least partially according to the inferred hierarchy and dependency relationships and including the aligned learning materials.

12. The non-transitory computer-readable medium of claim 11, wherein the learning materials include one or more of open courseware, massive open courses (MOOC), personal home pages, department home pages, and electronic books (e-books).

13. The non-transitory computer-readable medium of claim 11, wherein the structural information are extracted from learning materials metadata of the electronic learning materials and semi-structured format information included in the electronic learning materials.

14. The non-transitory computer-readable medium of claim 11, wherein the extracting structural information includes:

extracting lists from one or more of a syllabus, lecture notes, and a table of contents included in the electronic learning materials;

conducting a page-format analysis of one or more pages of the electronic learning materials;

detecting sequence borders based on one or more of line breaks, table cell borders, sentence borders, and specific punctuation; and

generating segmented term sequences from the lists and the page-format analysis, wherein one or more of the segmented term sequences are bounded by the detected sequence borders include positional information indicating a position in the electronic learning materials of the segmented term sequences.

15. The non-transitory computer-readable medium of claim 14, wherein the extracting knowledge points includes:

receiving the segmented term sequences having positional information indicating a position in the electronic learning materials of the segmented term sequence;

after the receiving the segmented term sequences, unifying abbreviations of the segmented term sequences;

after the unifying the abbreviations, constructing generalized suffix trees of the segmented term sequences;

after the constructing the generalized suffix trees, discovering repeated phrase instances of the segmented term sequences using the generalized suffix trees, wherein the phrase instances are limited by a particular maximum length;

after the discovering the repeated phrase instances, adjusting frequency of the discovered repeated phrases instances based on positions of phrase instances in the learning materials;

after the adjusting the frequency, measuring a cohesion and a separation of the segmented term sequences;

after the measuring the cohesion and the separation, removing stop phrases from the segmented term sequences;

after the removing the stop phrases, generating candidate knowledge points from the segmented term sequences;

after the generating the candidate knowledge points, calculating weights of the candidate knowledge points based on the adjusted frequency;

after the calculating the weights, analyzing appearance positions of the candidate knowledge points;

after the analyzing the appearance positions, constructing a hierarchy of knowledge points based on the appearance positions; and

after the constructing the hierarchy, presenting coverage overview of the learning materials.

16. The non-transitory computer-readable medium of claim 15, wherein:

the cohesion is measured according to a mutual information cohesion metric;

the separation is measured according to an accessor variety separation metric;

the abbreviations are unified according to a principal component analysis or a singular value decomposition; and

the weights are calculated according to a number of appearances and an authority of a particular learning material in which the segmented term sequences appear.

17. The non-transitory computer-readable medium of claim 15, wherein a hierarchy level of each of the knowledge points is based on a level of granularity of the knowledge points in the learning materials.

19. The non-transitory computer-readable medium of claim 11, wherein the aligning one or more of the knowledge points includes:

precisely locating each of the knowledge points in the learning materials based on granularity of the knowledge point; and

comparing and ranking the learning materials aligned with specific knowledge points.

20. The non-transitory computer-readable medium of claim 19, wherein the learning materials are compared and ranked according to a linear combination of a first weight multiplied by a knowledge point score, a second weight multiplied by a general score, and a third weight multiplied by a factor.

18. A non-transitory computer-readable medium having encoded therein programming code executable by one or more processors to perform or control performance of operations comprising:

deciding the relative positions between two knowledge points in the learning materials;

recommending immediate dependency relationships based on distance between the knowledge points in the hierarchy of knowledge points; and

generating a knowledge point map representative of the dependency relationships.