Understanding User Interactions with Podcast Recommendations Delivered Via Voice Lo Longqi Yang Michael Sobolev Christina Tsangouri Deborah Estrin Funders: 1
Emerging Voice Interfaces for Content Consumption 2
Diverse Content Delivered Through Voice Interfaces 3
Visual Interface Voice Interface Hi Hi, Al Alice, e, here ere are re the e ep episodes es rec recommen ended ed to Episode #1 you ou – nud nudge from the ted radio hour ur , , … Episode #2 Episode #3 Episode #4 Play Pl y the e ev evolutio ion of artif ific icia ial in intel ellig igen ence e from the e fres esh air ir … 4
What we may expect Users may spend longer time considering • each recommendation. Users may explore recommendation lists less. • Users may less likely choose items ranked • lower in recommendation lists. How much worse it is? Does it matter? 5
This Work A A be between-su subject ect r random omized ed c con ontrol ol st study y co comparing Voice and Visual interfaces Ef Effici cien ency Ex Exploration Choic Ch ice 6
Study Design Independent variable (Factor): In Interface Gr Group p A Gr Group p B Same list of podcast recommendations 7
Study Design Independent variable (Factor): In Interface Step 1 : St Browse a long list of episodes (1-N). • Br Making g a co committed ch choi oice: ce: typing • in the index number & listen to the episode for at least 5 minutes Gr Group p A Group Gr p B Same list of podcast recommendations 8
Study Design Independent variable (Factor): In Interface Step 1 : St Step 1 : St • Lis Listen n to a long list of episodes (1-N). Browse a long list of episodes (1-N). • Br • Making g a co committed ch choi oice: ce: typing in Making g a co committed ch choi oice: ce: typing • the index number & listen to the in the index number & listen to the episode for at least 5 minutes episode for at least 5 minutes Gr Group p A Group Gr p B Same list of podcast recommendations 9
Study Design Independent variable (Factor): In Interface St Step 1 : Step 1 : St • Listen Lis n to a long list of episodes (1-N). Browse a long list of episodes (1-N). • Br • Making g a co committed ch choi oice: ce: typing in Making g a co committed ch choi oice: ce: typing • the index number & listen to the in the index number & listen to the episode for at least 5 minutes episode for at least 5 minutes St Step 2 : St Step 2 : • Listen to the episode for at least 5 Listen to the episode for at least 5 • minutes minutes Gr Group p A Group Gr p B Same list of podcast recommendations 10
Study Design Independent variable (Factor): In Interface St Step 1 : Step 1 : St • Listen Lis n to a long list of episodes (1-N). Browse a long list of episodes (1-N). • Br • Making g a co committed ch choi oice: ce: typing in Making g a co committed ch choi oice: ce: typing • the index number & listen to the in the index number & listen to the episode for at least 5 minutes episode for at least 5 minutes St Step 2 : St Step 2 : • Listen to the episode for at least 5 Listen to the episode for at least 5 • minutes minutes Group Gr p A Group Gr p B Step 3 : St St Step 3 : A user satisfaction survey • A user satisfaction survey • Same list of podcast recommendations 11
Study Design Independent variable (Factor): In Interface 100 50 50 Group Gr p A Group Gr p B IT ITunes trendi ding g po podc dcasts 12
Efficiency The number of recommendations considered per minute Vi Visual Vo Voice 6 54 13
Exploration The maximum index number of the recommendations considered Vi Visual Vo Voice 15.3 47.8* 14
Choice: Lower-ranked Items The index number of the chosen episode Vi Visual Vo Voice 7.3 47.8 15
Choice: Lower-ranked Items preferred item stop (choice) 16
Main Implications 1/9 efficiency 1/ 1/ 1/3 exploration 1/6 chosen episode rank 1/ 17
Main Implications 1/9 efficiency Better navigation techniques 1/ 1/ 1/3 exploration 1/6 chosen episode rank 1/ 18
Main Implications 1/9 efficiency Better navigation techniques 1/ 1/ 1/3 exploration Adaptive and diverse recommendations at top ranks 1/6 chosen episode rank 1/ 19
Please refer to our paper for more experimental details and results! Pl Lo Longqi Yang Ph.D. candidate Computer Science, Cornell Tech, Cornell University Email: ylongqi@cs.cornell.edu Web: bit.ly/longqi Twitter: @ylongqi Connected Experiences Lab Small Data Lab http://cx.jacobs.cornell.edu/ http://smalldata.io/ Funders: 20
Recommend
More recommend