feature-engineering-studio · moiren · Sep 22, 2016 · charles-lang · Sep 23, 2016 · moiren
diff --git a/Problem Statement_Yaoli_talk and engagement.txt b/Problem Statement_Yaoli_talk and engagement.txt
@@ -0,0 +1 @@
+Problem Statement_Yaoli_talk and engagementVersion One:My problem is to develop a model to discover the attention pattern(fixation duration) of the visual features regarding the speaker and camera shooting in the ted talk videos that predict audience preference(scoring 1-5 on self-reported attitudes questions: interested, agree, engaging, informative). And my educational goal is to improve audience preference through manipulating attention-related features found in the talk video. My objective is to find the attention pattern that best predict audience preference. Version Two:My problem is to develop a model to discover the visual and audio features regarding the speaker and camera shooting in the ted talk videos that predict the attention pattern(fixation duration and location). My educational goal is to improve audience attention through manipulating visual and audio content features found in the talk video. My objective is to find the visual and audio features that best predict audience attention.I have collected a dataset of 29 participants�� fixation of two ted talk videos and their responses to 4 attitude questions. Currently my priorities are 1) to determine the granularity level of the features I��m coding and entering into the model. For example, how fine-grained should I code the gestures of the speaker, in level of general type(beats, deictics, iconic, metaphoric) or specific shapes and space(in central/peripheral space, high/low span, long/short stroke); 2) to determine the cut-off for defining a fixation(500ms from previous literature); 3) to normalize fixation duration based on total length of the video or specific visual features.See fake dataset in the attached excel.

diff --git a/fakedataset_Yaoli_talk and engagement.csv b/fakedataset_Yaoli_talk and engagement.csv
@@ -0,0 +1 @@
+,,versionOne,,,,,,,,,,,participantNo,participantGender,videoNo,speakerGender,fixation on gesture1(ms),fixation on gesture2(ms),fixation on speaker face(ms),fixation on slides(ms),fixation on other objects in the scene(ms),aggregated preference score of 4 attitude questions,,,,1,F,A,F,500,2000,20000,20000,100,10,,,,1,F,B,M,4000,20,10000,500,30000,18,,,,2,M,A,F,40,8000,40000,10000,18000,7,,,,2,M,B,M,1000,800,2000,40,2000,15,,,,,,,,,,,,,,,,,,,versionTwo,,,,,,,,,,,participantNo,participantGender,videoNo,speakerGender,gesture1Counts or Duration,gesture2Counts or Duration,speakerfaceDuration(ms),slidesDuration(ms),storyinspeech(describe)Duration(ms),reasoninginspeech(reason)Duration(ms),cameraZoominDuration(ms),cameraOnAudience(ms),aggreagted fixation time(ms)/counts ,aggreagted fixation time/counts on object11,F,A,F,,,,,,,,,,1,F,B,M,,,,,,,,,,2,M,A,F,,,,,,,,,,2,M,B,M,,,,,,,,,,
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		Problem Statement_Yaoli_talk and engagementVersion One:My problem is to develop a model to discover the attention pattern(fixation duration) of the visual features regarding the speaker and camera shooting in the ted talk videos that predict audience preference(scoring 1-5 on self-reported attitudes questions: interested, agree, engaging, informative). And my educational goal is to improve audience preference through manipulating attention-related features found in the talk video. My objective is to find the attention pattern that best predict audience preference. Version Two:My problem is to develop a model to discover the visual and audio features regarding the speaker and camera shooting in the ted talk videos that predict the attention pattern(fixation duration and location). My educational goal is to improve audience attention through manipulating visual and audio content features found in the talk video. My objective is to find the visual and audio features that best predict audience attention.I have collected a dataset of 29 participants�� fixation of two ted talk videos and their responses to 4 attitude questions. Currently my priorities are 1) to determine the granularity level of the features I��m coding and entering into the model. For example, how fine-grained should I code the gestures of the speaker, in level of general type(beats, deictics, iconic, metaphoric) or specific shapes and space(in central/peripheral space, high/low span, long/short stroke); 2) to determine the cut-off for defining a fixation(500ms from previous literature); 3) to normalize fixation duration based on total length of the video or specific visual features.See fake dataset in the attached excel.
Copy link Copy Markdown Contributor charles-lang Sep 23, 2016 Choose a reason for hiding this comment The reason will be displayed to describe this comment to others. Learn more. This looks very interesting, have you coded the videos already? If not that will be a very time consuming activity. Copy link Copy Markdown Author moiren Sep 26, 2016 Choose a reason for hiding this comment The reason will be displayed to describe this comment to others. Learn more. Yes, I've coded the videos myself without inter-rater reliability.
Expand Down
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		,,versionOne,,,,,,,,,,,participantNo,participantGender,videoNo,speakerGender,fixation on gesture1(ms),fixation on gesture2(ms),fixation on speaker face(ms),fixation on slides(ms),fixation on other objects in the scene(ms),aggregated preference score of 4 attitude questions,,,,1,F,A,F,500,2000,20000,20000,100,10,,,,1,F,B,M,4000,20,10000,500,30000,18,,,,2,M,A,F,40,8000,40000,10000,18000,7,,,,2,M,B,M,1000,800,2000,40,2000,15,,,,,,,,,,,,,,,,,,,versionTwo,,,,,,,,,,,participantNo,participantGender,videoNo,speakerGender,gesture1Counts or Duration,gesture2Counts or Duration,speakerfaceDuration(ms),slidesDuration(ms),storyinspeech(describe)Duration(ms),reasoninginspeech(reason)Duration(ms),cameraZoominDuration(ms),cameraOnAudience(ms),aggreagted fixation time(ms)/counts ,aggreagted fixation time/counts on object11,F,A,F,,,,,,,,,,1,F,B,M,,,,,,,,,,2,M,A,F,,,,,,,,,,2,M,B,M,,,,,,,,,,
Expand Down