ࡱ> GF( %  vpathRhttp://www.cs.princeton.edu/ugradpgm/bse// 0DArialNew bb ,bhb0bz[ 0xbDWingdingsbb ,bhb0bz[ 0xb DTimes New Romanbhb0bz[ 0xb B .  @n?" dd@  @@``  PUV        0AA 3@8y FlO ʚ;Sk8ʚ;g4MdMdbz[ 0bppp@ <4ddddĶb 0b+b~0___PPT10 >___PPT9 ? %O ='Princeton University Course Recommender4Senior Project By Paul Simbi Advisor: Andrea LaPaugh5Z5 DescriptionRecommend courses using various Collaborative Filtering (CF) algorithms Predict the utility of courses to a particular student based on a database of user votes from other users.&HkHkWhy is this Important?^There are a wide variety of classes As the amount of choices (courses) grow very hard to evaluate every choice Help out freshman Aid in picking major Make advisor s job easierN$(#A$(#AWhy is this hard?XMultiple ways to figure out which user or item one is  close to. Depends on items (books, music, etc.) Is the data sparse Usually done with media Students and courses have & Prerequisites Majors Distribution requirements BSE vs. AB Courses change Professors Format New users Freshman have no historyBP9PPP:PPP PPB9:   Previous ApproachesTypically done on media Books, movies, music Amazon MovieLens Launch.com At Princeton Have recommended path for different majors. No Information filteringj-ZZ Z,ZZ- ,U 0[gU 0gk My ApproachInformation Filtering Majors, AB vs. BSE, prerequisites, etc. Eliminate unwanted courses Learn new user preferences (freshman) Quick registration Extracts useful data Present courses/subjects to a new user New user votes Enjoyment and difficulty of class(&<1(& <1  My ApproachFTest different algorithms Memory-Based Find set of users, neighbors, that are similar to the target user (user-based) Model-Based More probability based Estimate probabilities Item-Based Person is interested items similar to those he has liked before Better? Courses haven t been items in CFZ ZOZ Z.Z Z@ZZ!Z  3 .  @!  MethodologybModel-Based Bayesian Network Model Cluster Model Memory-Based User-based Item-based Group of usersL % % % %   MethodologyTweak Algorithms & Parameters Ex. Weight computation Correlation, Vector Similarity, etc Learning new users Best questions to ask new users Ex. Log Popularity * Entropy Information filteringv$=$ =    MethodologyExperiments Feedback from user Which recommendation they think is good His/her predicted vote Precision and Recall Mean Absolute Error (MAE)N ?/ ?/ ! Checkpoint (November 20th):Implement Model Based Bayesian Network Model Cluster Model&%%"End of the semester (Dec. 9-12)`Tweak the algorithms Figure out how much data needed Figure out how to get data Change Interface# Fall Back5Algorithms Concentrate on less Change less parameters& + +$SummaryPredicting courses a student Enjoyment Difficulty Collaborative & Information filtering Model-based Memory-Based Item-Based EvaluatejZZ&Z$Z Z&$  b` 3fffff3̙3f̙` ̙3f` ff3f` f33f3f` 3ffƍ` fff3` f33̙` 3f|>?" dZ@$?lKd@   l@  P`lA n?" dd@   @@``PT   @ ` `p>>     ` (   F  "TN  "n  0"   $0e0e BCDELF>A5%8c8c     ?1d0u0@Ty2 NP'p<'p@A)BCD|E?|b4"%>Ul @   `c"$  XN  "n"   0G"r B T??"  <\GH "`  T Click to edit Master title style! !*  6  "  RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S    ` ??"`>   @*   ` ??"`@a   @*   `T ??"\5  D*B  s *޽h ? 3fffff3̙3f̙ Capsules]  b     (   F p  "p   <I"@  ."   HxLG0* "pp  .~Z     # "  t"  B 6G" R r   Z??"L    <x+ "4 `   W#Click to edit Master subtitle style$ $   `R ??"`>   D*   `V ??"`@a   @*   `[ ??"`0  D*"   f_G0* ??"p   T Click to edit Master title style! !B  s *޽h ? 3fffff3̙3f̙ PL8(  L L 0 P    >*  L 0̼     @*  L 6Т _P   >*  L 6̩ _   @* H L 0޽h ? 3380___PPT10.zv$b 0$(  r"  S h p   r  S xi 4 `   H  0޽h ? 3fffff3̙3f̙80___PPT10.y $ b @ $(   r"   S `   r   S T  H   0޽h ? 3fffff3̙3f̙80___PPT10.y4$ b $(  r"  S 욉`   r  S   H  0޽h ? 3fffff3̙3f̙80___PPT10.y7$ b  $(   r"   S ̉`   r   S ͉  H   0޽h ? 3fffff3̙3f̙80___PPT10.y@n$ b $$(  $r" $ S `   r $ S   H $ 0޽h ? 3fffff3̙3f̙80___PPT10.y`sV$ b ($(  (r" ( S `   r ( S   H ( 0޽h ? 3fffff3̙3f̙80___PPT10.y$ b 0$(  0r" 0 S 0%`   r 0 S &  H 0 0޽h ? 3fffff3̙3f̙80___PPT10.Kz`o} b ,$(  ,r" , S 3`   r , S ,4  H , 0޽h ? 3fffff3̙3f̙___PPT10i.y`D?+D=' b= @B +} b 8$(  8r" 8 S >`   r 8 S ?  H 8 0޽h ? 3fffff3̙3f̙___PPT10i.Uz@c+D=' b= @B +} b 4$(  4r" 4 S D`   r 4 S E  H 4 0޽h ? 3fffff3̙3f̙___PPT10i.Sz+D=' b= @B +$ b <$(  <r" < S L`   r < S M  H < 0޽h ? 3fffff3̙3f̙80___PPT10.Vz^$ b @$(  @r" @ S dS`   r @ S 8T  H @ 0޽h ? 3fffff3̙3f̙80___PPT10.Vz$ b  D$(  Dr" D S hg`   r D S |W  H D 0޽h ? 3fffff3̙3f̙80___PPT10.Vz $ b 0H$(  Hr" H S 0;`   r H S <  H H 0޽h ? 3fffff3̙3f̙80___PPT10.Vz$rT0k- :=H?tACE$JG.OLQS V7X6? cZ% Oh+'0 px    ( 4@H(Princeton University Course Recommender Paul Simbin Capsulesin Paul Simbin16lMicrosoft PowerPointCou@@fUX@@`'y@@iV z\Grg    - @ !--'̙- @ !--'--$Sllmmnqtyy~qjc]XSPMLLKKKLLMPSX]cjq%y+058;<<====$<)<.;88B5K0T+\%cjpuz}}zupjc\T~KyBt8q.n)m$mlll--'3f--$ 9999--'--,$"%(,/2579:::--'. 3f@Arial-12 Princeton University Course !$$.Courier New((,bi-.@Arial-2 _ Recommenderd$,,.--. ff@Arial-2 j Senior Project    .-.@Arial-2   By Paul Simbit   ! .-.@Arial-2  Advisor: Andrea     .-.@Arial-2  LaPaugh.- ՜.+,D՜.+,8    On-screen ShowPrinceton UniversityrZhA Arial WingdingsTimes New Roman Capsules(Princeton University Course Recommender DescriptionWhy is this Important?Why is this hard?Previous Approaches My Approach My Approach Methodology Methodology MethodologyCheckpoint (November 20th) End of the semester (Dec. 9-12) Fall BackSummary  Fonts UsedDesign Template Slide Titles(PX _PID_HLINKSVersionA*http://www.cs.princeton.edu/ugradpgm/bse/"_Z Paul SimbiPaul Simbi  !"#$%&'()*+,-/012345789:;<=?@ABCDEHRoot Entry% dO)GDICFGDICCurrent User !b ;% X4Kl=>KSummaryInformationl=S=(l=<>Kl.%PowerPoint DocumentFGDICF(:F(GDICZDocumentSummaryInformation86