A Combined Method for E-Learning Ontology Population based on NLP and User Activity Analysis
Download as PPTX, PDF1 like1,330 views
The document outlines a combined method for enhancing e-learning ontology through natural language processing (NLP) and user activity analysis, specifically aimed at improving educational materials and tutor resources. It details the development of various ontologies for education and tests, the use of RDF for test conversion, and a user-friendly front-end system supported by analytics tools. Future work includes refining term extraction procedures and improving the quality of the ontology through better data processing and analysis methods.
A Combined Method for E-Learning Ontology Population based on NLP and User Activity Analysis
- 1. A Combined Method for E-Learning Ontology Population based on NLP and User Activity Analysis Dmitry Mouromtsev, Fedor Kozlov, Liubov Kovriguina and Olga Parkhimovich Laboratory ISST @ ITMO University, St. Petersburg, Russia Linked Learning meets LinkedUp: Learning and Education with the Web of Data @ ISWC 2014, Italy
- 2. Introduction ● Use semantics to make education materials reusable and flexible, ● Different datasets in one e-learning system, ● We need to provide tools for tutors to improve their courses.
- 3. Use students to improve your course!
- 4. Goal ● To develop the ontologies, ● To convert tests from XML to RDF, ● To map tests with subject terms via NLP algorithms, ● To gather user’s statistics and implement analysis module.
- 5. ECOLE: Front-end ● Ontology-based e-learning system, ● User friendly interface, ● Based on Django framework.
- 6. ECOLE: Back-end ● Collection of educational materials from different open resources (Dbpedia, BNB), ● Analytics tools, ● Based on the Information Workbench platform.
- 7. The ontology of education resources ● Extends AIISO ontology for education process and structure, ● Uses BIBO for bibliographic resources, ● Uses MA-ONT for media resources.
- 8. The ontology of tests ● Developed to describe the content of tests, ● Extends top-level ontology of the system, ● 12 classes, ● 10 object properties, ● 6 datatype properties, ● http://purl.org/ailab/testontology.
- 9. The ontology of student activity ● Developed to store information about the student's learning process and results, ● Uses FOAF, ● Uses the ontology of test, ● 10 classes, ● 15 object properties, ● 5 datatype properties, ● http://purl.org/ailab/learningresults .
- 10. Lemmatization ● Extracts lemmas from subject term labels, ● Uses NLP procedures and dictionaries to generate lemmas, ● Stores lemmas in triples.
- 11. Terms extraction
- 12. New System Terms ● If the extracted word sequence doesn't match any of the system terms, it is included as a new system term, ● New system terms are validated via SPARQL queries to Dbpedia.
- 13. NLP Algorithm Dictionary Patterns Set Tokenization Morphological and Semantic Tagging Extracting Candidate Terms Producing the Canonical Form of the Term + Lemma(s) ● Uses Part-of-Speech patterns ● Grammatical agreement is specified in the patterns ● Words+Lemmas+Morphology+Some semantic tags ● Superlemmas (a common lemma is generated for lexical variants) ● Derivation paradigms (a common lemma is generated for words with the same root)
- 14. Implementation: Test parsers ● Converts tests of the course from XML to RDF, ● Uses Information Workbench XMLProvider to automatically update, ● Describes mapping using XPath functions.
- 15. Implementation: NLP module ● Uses dictionaries in NooJ format: <LEMMA>+<PART OF SPEECH TAG>+ <INFLECTIONAL PARADIGM>+<OTHER ANNOTATIONS> air,N+FLX=TABLE Michelson,N+ProperName ● English NooJ resources are reused, Russian lexical resources are original, ● A separate procedure implemented in Python is launched for lemmatization and extraction of the terms.
- 17. Hidden subject terms A ladder is 5m long. How far from the base of a wall should it be placed if it is to reach 4m up the wall? Nothing to extract!!! Hidden subject terms: Pythagorean Theorem, Hypotenuse, Сathetus…...
- 18. User statistics collection ● Collects user's activity on front-end, ● Creates triples, ● Analyzes user's answers on tests, ● Builds rating of the terms, which caused difficulties for students.
- 19. Conclusion ● The ontologies of tests and student activity have been developed, ● The tasks of the test have been linked with system terms, ● The statistics gathering module has been developed, ● The system provides rating of the terms, which caused difficulties for students.
- 20. Future Work ● Improve term extraction procedure by adding parallel texts of tasks, ● Process units of measure in tasks to predict “hidden terms”, ● Use relations between subject terms to improve the quality of term extraction procedure, ● Refine term knowledge rating by replacing it by the proper ranking formula.
- 21. Thank you!!! The front-end of the e-learning system: http://ecole.ifmo.ru Example of subject terms analytics for module "Interference and Coherence": http://openedu.ifmo.ru:8888/resource/Phisics:m_InterferenceAnd Coherence?analytic=1 The source code: https://github.com/ailabitmo/linked-learning-solution
Editor's Notes
- #5: To develop the ontology of tests, To develop the ontology of student activity, To convert tests from XML to RDF, To extract terms from tasks using morpho-syntactic patterns, To link tasks to the lecture terms via extracted candidate terms, To validate candidate terms via DBpedia, To develop statistics gathering module, To create pages to show statistics of students' answers to the tests.
- #8: ??? example of indirect links ???
- #9: The main purpose of the developed ontology is to represent structural units of a test and provide automatic task matching by defining semantic relations between tasks and terms
- #10: Two top-level ontologies have been used for its development: ontology of test, as described above, and FOAF ontology that describes people and relationships between them. The e-learning system uses the ontology of tests and answers given by the user to build a list of terms that the user knows.
- #12: Collects tasks of the course using SPARQL queries, Forms the plain text content for each task using the information about questions and answers of the task, Launches NLP procedures for the content of the task, The result data contain a canonical form and lemma(s) for the candidate term, Search terms in the system to link them with candidate terms, System terms and candidate terms are linked, if they have the same lemma(s).
- #14: NLP algorithm has plain text of the task in the input. This plain text is tokenized and tokens are tagged basing on lexical resources that are stored in the dictionary. Candidate terms are extracted with part-of speech patterns. About a dozen patterns is used for term extraction. Russian patterns have more constraints, because it is necessary to specify grammatical agreement in the pattern. For each candidate term a canonical form is produced to validate this term in the DBpedia and lemma(s) to map the term to system terms. //////The canonical form corresponds to the form of the headword of this term in dictionaries. /////////// Uses Part-of-Speech patterns to extract terminology For the Russian language grammatical agreement is specified in the patterns, Example: <N+nom+sg><N+gen+sg+Prop> to extract Russian terms like "бипризма Френеля" ("Fresnel biprism"), Superlemmas allow to link all variants via a canonical form and store them in one equivalence class (e.g, “rectangular Cartesian coordinate system” and “RCCS” are matched via the same query), Derivational paradigms are described, which allows to produce a common lemma for words belonging e.g. to different parts of speech (that is, "coplanar" and "coplanarity" will be lemmatized as "coplanar".
- #16: We use lexical resources built in/generated in NooJ linguistic environment. Dictionaries are written in NooJ format *dic: <LEMMA>+<PART OF SPEECH TAG>+<INFLECTIONAL PARADIGM>+<OTHER ANNOTATIONS> (red - necessary, yellow - optional). Example: air,N+FLX=TABLE, Michelson,N+ProperName (red & orange - necessary elements for the dictionary to be valid, green & yellow - optional). When we have specified the inflectional paradigm, it is possible to generate all word forms with corresponding lemmas and morphological and semantic tagging. //aperture,aperture,N+FLX=TABLE+s //apertures,aperture,N+FLX=TABLE+p For the demo version in English standard NooJ lexical resources were reused, Russian lexical resources are original. After tagging a Python script runs to lemmatize input tokens and extract terms.
- #17: Percent of linked tasks - a good result; it shows that 95% of tasks are suitable to test the student’s knowledge, the test set is generally valid. Number of different candidate terms - precision needs improvement! There are 7.5 candidate terms per test in the average, that is, certainly, isn’t true (content of the physics test module has 1-2 terms per task (NB! they may repeat in the task). This indexes to the number of false candidates equal to the number of true candidates and shows the necessity of the validation procedure. Number of manually extracted different terms - a “gold standard”. Thus, we have a 1-2 terms circulating in the task. It seems, that the amount of terms in the tasks is domain-dependent. Percent of system terms, matched by candidate terms - needs improvement. Improvements can be done by adding periphrases of POS-patterns. Tasks testing some system terms have only “hidden terms”.
- #19: The data about number of correct/incorrect user's answers allow to compute the knowledge rating for any of the system terms. Using this rating, teachers can determine which terms of the course students know worst of all.
- #20: The developed modules for the e-learning system provide teachers with a tool to maintain relevance and high quality of existing knowledge assessment modules. The system provides rating of the terms, which caused difficulties for students. Based on this rating, teachers can change theoretical material of the course by improving description of certain terms and add proper tasks.