Affective Computing and Intelligent Interaction Ma. Mercedes T. Rodrigo Ateneo Laboratory for the Learning Sciences Department of Information Systems and Computer Science Ateneo de Manila Univeristy
Page 2 Affective computing Computing that relates to, arises from or deliberately influences emotion - Picard, 1997, Affective Computing
Page 3 Affective computing Emotion recognition Emotion expression Intelligent response to emotion
Page 4 Significance: Towards more humane interfaces How can we enable computers to better serve people’s needs--adapting to you, vs. treating you like some fictionalized ideal user, and recognizing that humans are powerfully influenced by emotion, even when they are not showing any emotion? - Picard, 2003
Page 5 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 6 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 7 Methods: Aplusix
Page 8 Methods: Ecolab / MEcolab
Page 9 Methods: Cognitive Tutor
Page 10 Methods: BlueJ
Page 11 Methods: The Incredible Machine
Page 12 Methods: Math Blaster
Page 13 SQL/EER Tutors
Page 14 SimStudent
Page 15 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 16 Biometrics instruments
Page 17 Biometrics instruments
Page 18 Biometrics instruments
Page 19 Sample Log: Brainfingers [Header V2035] 7/31/2009 7:49:05 PM UserFile = C:\Nia Data\\__ usr [Data] Sample,Event,GlanceMagJs,GlanceDirJs,A1Js,A2Js,A3Js,B1Js,B2Js,B3Js,MuscleJs 1,0,0.1,0.065,0.0195, , E-07,0.5815,0.5311,0.6048,0.6782,0.7665,0.7074,0 1,0,-0.029,-0.122,0.021,0.2056, E-07,0.5774,0.5251,0.5892,0.6723,0.7598,0.7125,0 1,0,0.015,-0.167,0.0205, , E-07,0.5743,0.5187,0.5737,0.6683,0.7534,0.7157,0 1,0, , ,0.0205, , E-07,0.573,0.5118,0.5596,0.6656,0.7468,0.7159,0 1,0,-0.285,-0.163,0.0205, , E-07,0.5733,0.5043,0.5477,0.6622,0.7398,0.7168,0 1,0, ,-0.206,0.022, , E-07,0.5745,0.4966,0.5371,0.6567,0.7321,0.7221,0 1,0,-0.125,-0.158,0.0225, , E-07,0.5772,0.4885,0.5268,0.6483,0.7242,0.7262,0
Page 20 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 21 Sample Log: Aplusix %;ACTIONS;#Date=1/16/2007#Heure=14:57:59;#TypeProbleme=TpbDevelopper 0;0.0;structure;();0;();();();();();(); 2 derriere);rien;;N1; 2 derriere);rien;V1;N1; 6;1.5;BackSpace;();1;?;();(dedans);rien;V-;N-; 7;2.0;-;();1;-?;();(0 dedans);rien;V-;N-; 8;3.7;4;();1;-4;();(0 0 derriere);rien;V0;S0; 9;0.2;9;();1;-49;();(0 1 derriere);rien;V0;S0; 10;2.7;x;();1;-49x;();(0 1 derriere);rien;V0;S0; 1 1 dedans);rien;V-;N-;
Page 22 Sample Log: Ecolab New Activity Toolbar ButtonClick0 6 Activity 16 Activity Chosen: Food 46 Suggested Help 06 Suggested Challenge 16 Challenge Accepted18 ViewWeb change13 ViewWeb change14 ActionShow19
Page 23 Sample Log: Scatterplot Tutor *000:03:781 READY. *000:59:503 APPLY-ACTION WINDOW; ALGEBRA-2-TRANSLATOR::VARIABLE-TYPE-MODEL, CONTEXT; SPLOT-DB-C , SELECTIONS; (|var-1val-1|), ACTION; SUBSTITUTE-TEXT-INTO-BLANK, INPUT; ("Numerical"),. *000:59:503 UPDATE-P-KNOW META; META-VALUING-NUM-FEATURES, PRODUCTION; (CHOOSE-VAR-TYPE-NUM MIDSCH-VARIABLE-TYPING), SUCCESS?; T, P-KNOW; ,..
Page 24 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 25 consolidation analysis Methods
Page 26 Methods
Page 27 Methods
Page 28 analysis Methods
Page 29 analysis consolidation analysis Methods
Affective states
Page 31 Boredom
Page 32 Confusion
Page 33 Delight
Page 34 Frustration
Page 35 Engaged concentration
Page 36 Others Neutrality Surprise
Behaviors
Page 38 On-task, solitary
Page 39 On-task, giving and receiving answers
Page 40 Other on-task conversation (probably)
Page 41 Off-task solitary
Page 42 Others Off-task, conversation Inactive Gaming the system
Page 43 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 44 Talk to me %;ACTIONS;#Date=1/16/2007#Heure=14:57:59;#TypeProbleme=TpbDevelopper 0;0.0;structure;();0;();();();();();(); 2 derriere);rien;;N1; 2 derriere);rien;V1;N1; 6;1.5;BackSpace;();1;?;();(dedans);rien;V-;N-; 7;2.0;-;();1;-?;();(0 dedans);rien;V-;N-; 8;3.7;4;();1;-4;();(0 0 derriere);rien;V0;S0; 9;0.2;9;();1;-49;();(0 1 derriere);rien;V0;S0; 10;2.7;x;();1;-49x;();(0 1 derriere);rien;V0;S0; 1 1 dedans);rien;V-;N-;
The problem: 5(-9x+7)+3(4x-3)
Step #Text typed by student 21-45x 22-45x=? 23-45x 24-45x+? 25-45x x x+35+? 28-45x x x+35+12x 31-45x+35+12x-? 32-45x+35+12x-9
The problem: 8x 2 -2x+6-(-5x 2 +8x+3)
Page 48 Step #Text typed by student 563c 564ch 565chr 566chri 567chris 568christ 569christi 570christin 571christine 572christine+? 573christine+c 574christine+cy 575christine+cyr
Step #Text typed by student 576christine+cyri 577christine+cyril 578christine+cyril=? 579christine+cyril=a 580christine+cyril=ab 581christine+cyril=abi 582christine+cyril=abig 583christine+cyril=abiga 584christine+cyril=abigai 585christine+cyril=abigail
Page 50 I’m not a math genius but I’m pretty sure that 8x 2 -2x+6-(-5x 2 +8x+3) != christine+cyril=abigail
Page 51 Analysis methods Clean the data Define the different features Distill new features Define desired range of values Select an appropriate statistical test or data mining algorithm Validate the findings
Page 52 Analysis techniques Statistical methods Data mining techniques
Page 53 Persistence of affective states Regardless of software, boredom tends to persist Affect and behavior detection
Page 54 Affect and behavior Students who attempt the most difficult problems experience flow the most Students who try the lowest levels experience more boredom and confusion. Students who take the longest time in solving the problems experience confusion the most Students who take the shortest time experience confusion the least. Students who use the most number of steps to solve a problem experience confusion and boredom the most. Students who take the least number of steps experience more flow.
Page 55 Carelessness Creation of models that generalize across different datasets and countries Students who are bored and confused are less likely to be careless Students who are engaged are more likely to be careless
Page 56 Novice programmer behaviors Students who are persistently confused do worse on the midterm Students who are able to resolve their confusion do better Students with low error quotients do better on the midterm
Page 57 Aplusix, Scatterplot, Ecolab, BlueJ, etc. Student Interaction Logs Observation Logs Analysis Affect detectors Behavior detectors Novice programmer errors Interventions Intelligent agents Improved error messages Sensor data
Page 58 Example
Page 59 New version
Page 60 Affect-sensitive game
Page 61 BlueJ Browser
Page 62 BlueJ Browser
Page 63 EDM Workbench
Page 64 Closing the loop We continue to create new models and integrate the models with software We still have a long way to go
My thanks Ryan Baker Jessica Sugay Thor Collin Andallaza Rina Joy Jimenez Jason King Li Tricia Monsod Diane Lee Minmin Lagud Sweet San Pedro Thomas Dy And all the “cast of thousands” who constitute the ALLS
Thank you! Questions?