Student Modeling with Automatic Knowledge Component Extraction for Adaptive Textbooks
- 2. Contoso Pharmaceuticals OnlineTextbooks Trust Integer convallis suscipit ante eu varius. Morbi a purus dolor. Cost Suspendisse sit amet ipsum varius finibus justo viverra blandit. 2 Optional / Mandatory reading components Flip Courses Blended Courses MOOCs
- 3. Contoso Pharmaceuticals ResearchonReadingactivities Students who completed optional readings did better • Ut congue quis tortor eget sodales. Nulla a erat eget nunc hendrerit ultrices eu nec nulla. • Donec viverra leo aliquet, auctor quam id, convallis orci. 3 Students indicate reading as optimal
- 4. Contoso Pharmaceuticals StudentModelingFramework State-of-the-art student modeling frameworks are based on traditional ITS Main focus of ITS is to increase student performance through practice activities Basic architecture of an ITS 4
- 5. Contoso Pharmaceuticals TraditionalStudentModelingFramework State-of-the-art student modeling frameworks are based on traditional ITS Main focus of ITS is to increase student performance through practice activities Ignore student reading behavior Basic architecture of an ITS1 5
- 6. Contoso Pharmaceuticals Student models that considerreading • Variant of Bayesian Knowledge Tracing • Comprehension Factor Analysis 6 Reading behavior helps in inferring Student’s knowledge state
- 7. Contoso Pharmaceuticals State of the art reading models for performance prediction • Variant of Bayesian Knowledge Tracing • Comprehension Factor Analysis 7 Subject Experts Knowledge Components/ SKILLS
- 10. Contoso Pharmaceuticals ACETechniques 10 Noun Phrase Extraction WikipediaTitle based Filtering Topic models – LDA Title as a concept Topic Based Concept ExtractionPhrase based Concept Extraction
- 11. Contoso Pharmaceuticals ACETechniques 11 Noun Phrase Extraction
- 12. Contoso Pharmaceuticals ACETechniques 12 WikipediaTitle based KCs Wikipedia Title list Textbook N grams Top N Keyphrases
- 14. Contoso Pharmaceuticals 14 Reading Activities KCA KCB KCA KCA KCB KCA R = 1 R = 3 Reading Opportunity every page they spent substantial amount of time KC – Knowledge components
- 15. Contoso Pharmaceuticals log 𝑝𝑖𝑗 1 − 𝑝𝑖𝑗 = 𝛼𝑖 + 𝑘 𝛽 𝑘 𝑄 𝑘𝑗 + 𝑘 𝑄 𝑘𝑗 ( 𝜇 𝑘 𝑆𝑖𝑘 + 𝜌 𝑘 𝐹𝑖𝑘 + 𝛾 𝑘 𝑅𝑖𝑘 where, 𝑝𝑖𝑗 is probability of success of student 𝑖 on item 𝑗 𝑖 is a student, 𝑗 is a step and 𝑘 is a Skill 𝛼𝑖 is a coefficient associated with student 𝑖 represents the proficiency of student 𝑖. 𝑄 is a Qmatrix 𝑄 𝑘𝑗 is Qmatrix cell associated with item 𝑗 and Skill 𝑘 𝛽 𝑘, 𝜇 𝑘, 𝜌 𝑘 𝑎𝑛𝑑 𝛾 𝑘 are coefficients associated with skill 𝑘 𝑆𝑖𝑘 𝑎𝑛𝑑 𝐹𝑖𝑘 as number of success and failure attempts respectively of student 𝑖 on skill 𝑘 𝑅𝑖𝑘 is reading opportunities of student 𝑖 on skill 𝑘 Reading Activity 15 Reading behavior Classic PFA Reading Opportunities Skill connected with reading skill hardness Performance on quiz
- 16. Contoso Pharmaceuticals Student Performance on PredictionTask ACE METHOD No of Concepts Accuracy Title 25 48.43 LDA 200 56.34 Noun Phrase 619 60.32 Wikipedia 521 61.34 10 fold cross validation on CFM model 16 Surprises: No of concepts do not have a substantial effect on student models NP and Wikipedia extracted too many concepts which break the assumption of independence Topics are under representative of concepts Key-phrases are over representative and bring about too many dependent concepts in paly
- 17. Contoso Pharmaceuticals Future work page 17 • CombineTopic based concepts + Key phrase based concepts • Compare the models with expert annotated concepts • 30% of reading activities were skimming • Understand how pre-requisites and concepts can be differentiated