1 / 18

Analyzing forum interactions on a MOOC

Analyzing forum interactions on a MOOC. Jonathan Huang. ml-class forum. MOOC Forums. Forum data is something that we want to study. And we have a lot of it. But how are we going to read through 34,000 posts?!?!?. LDA Output. .1 algebra .06 computers .02 programming …. 40%. 1.

magar
Download Presentation

Analyzing forum interactions on a MOOC

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Analyzing forum interactions on a MOOC Jonathan Huang

  2. ml-class forum

  3. MOOC Forums Forum data is something that we want to study. And we have a lot of it. But how are we going to read through 34,000 posts?!?!?

  4. LDA Output .1 algebra .06 computers .02 programming … 40% 1 Like baseball after sabermetrics, like Wall Street in the 1970s, sportsgambling over the past decade has undergone a quantitative revolution. Nearly every successful sports bettor in the world now uses some form of computer model to assist in the handicapping of sporting events. Like their brethren inside hedge funds, these gamblers are known as quants. Like the advanced tradingsystems operating on Wall Street, the models used by this technologically adroit breed of sports bettor are sometimes called black boxes. Their models (and their identities) are shrouded in secrecy. Their algorithms are proprietary. [From Scott Eden, ESPN] .12 economics .1 trading .08 hedge fund … 15% 2 .09 basketball .08 match .04 running … 45% 3 Topic-word probability matrix Per document topic probability vector Per document topic-word assignment

  5. Topics explicitly covered by Andrew Ng “SVMs” “PCA” “Recommender Systems” “Neural Networks”

  6. MOOC topics “Certification” “Discussion and Study Groups” “MOOC” “Video Lectures”

  7. Social topics “Self Introductions” “Español” “Andrew Ng Love”

  8. Technical support topics “Code submission” “Vectorization” “Octave”

  9. Machine learning beyond the classroom “Cool Applications of ML” “Going on”

  10. Thread level analysis (Thread #11) ----------------------------- Hello, my name is Bob. I am a college graduate with a degree in applied mathematics. I would like to apply to a computer science program next year. My GPA was not so good. I would like to use this course to boost my application. Suppose I earn “top marks” on this “free course”, how might I put it into my graduate application? ----------------------------- I think it would be of minimal value because what we are going to receive is a Statement of Accomplishment, signed by one of the instructors and definitely not a Stanford certificate. It may be helpful but not for sure, depends on what one has actually learned in this course. ----------------------------- I don't think it would count for much in grad school applications. What will help you are excellent recommendation letters, published papers and good GRE scores. However it wouldn't hurt your application either and you will probably learn stuff relevant to CS grad school (especially if you specialize in machine learning!). ----------------------------- This should not be your main goal, because it is possible you got disappointed with future recognition. On my point of view, the main goal should be to learn about machine learning. You will be able to have recognition once you apply your knowledge you got in this course to develop something cool. ----------------------------- Courses Attended: Machine Learning (Online) Institution/School: Stanford Qualification: Statement of Accomplishment (or whatever is written on the statement of accomplishment or how you performed) Well i don't know for you guys, but i am taking the course seriously and if i happen to be successful, i believe i should be proud of it. And any institution would recognize it you matter how little. After all people take online masters courses and are awarded certificates. The only problem is that it cannot be proven that you genuinely attended the course, but note the institution might decide to test you on the course in an interview and if you perform well its going to be definitely an added advantage but if you perform poorly, you might even be considered as a fraudster. ----------------------------- I suppose that adding it to a CV as part of let me say Continuous Learning under Education would be helpful in case you do not have much to add. Online Course “Machine Learning” by Prof. Andrew Ng Organised by Stanford University probably would be fine, right?

  11. Per-document topics 10 21 Probability of topic 15 46 42 Five most prominent topics in Thread 11

  12. Word-topic Assignments

  13. Topic Activity Over Time 10/1/2011 12/31/2011

  14. Topic Activity Over Time 10/1/2011 12/31/2011

  15. Topic popularity by votes and views Andrew Ng Self Introduction Linear Algebra Linear Algebra Number of votes by topic Self Introduction Andrew Ng Self Introduction Andrew Ng Linear Algebra View/vote ratio by topic Number of views by topic

  16. For each forum question q, estimate: Increase in probability of transitioning to correct response given that a user viewed question q Forum visits (focusing only one “cluster” of incorrect responses at a time) Did forums help? Question: Did forum views help users on the homework assignments? Experimental setup: “incorrect” submission Final “correct” submission “incorrect” submission Submission timeline

  17. Most common incorrect submission Correct solution function J = computeCost(X, y, theta) %COMPUTECOST Compute cost for linear regression m = length(y); J = 2; for i= 1:m;    J=J+(theta(1,1)+theta(2, 1).*X(i)-y(i))^2; %Implementation of the cost function.   end J=(1/(2*m)).*J function J = computeCost(X, y, theta) %COMPUTECOST Compute cost for linear regression J = sum((X * theta - y) .^ 2) / (2*m); Did forums help? Question: 911, with 30 views and 2017 not viewed.  +Fraction transitioned to correct if viewed: 0.733  +Fraction transitioned to correct if not viewed: 0.349  +Fraction improvement by viewing: 2.104 Question: 830, with 98 views and 1949 not viewed.  +Fraction transitioned to correct if viewed: 0.643  +Fraction transitioned to correct if not viewed: 0.34  +Fraction improvement by viewing: 1.893 Question: 1348, with 32 views and 2015 not viewed.  +Fraction transitioned to correct if viewed: 0.594  +Fraction transitioned to correct if not viewed: 0.35  +Fraction improvement by viewing: 1.695 Question: 805, with 39 views and 2008 not viewed.  +Fraction transitioned to correct if viewed: 0.59  +Fraction transitioned to correct if not viewed: 0.35  +Fraction improvement by viewing: 1.687 Question: 728, with 89 views and 1958 not viewed.  +Fraction transitioned to correct if viewed: 0.573  +Fraction transitioned to correct if not viewed: 0.344  +Fraction improvement by viewing: 1.665 Question: 1223, with 26 views and 2021 not viewed.  +Fraction transitioned to correct if viewed: 0.577  +Fraction transitioned to correct if not viewed: 0.351  +Fraction improvement by viewing: 1.642 Question: 1256, with 71 views and 1976 not viewed.  +Fraction transitioned to correct if viewed: 0.563  +Fraction transitioned to correct if not viewed: 0.347  +Fraction improvement by viewing: 1.625 Question: 932, with 59 views and 1988 not viewed.  +Fraction transitioned to correct if viewed: 0.542  +Fraction transitioned to correct if not viewed: 0.349  +Fraction improvement by viewing: 1.556

More Related