Understanding a and R Recommending Po Podcast Content Longqi - PowerPoint PPT Presentation

Understanding a and R Recommending Po Podcast Content Longqi Yang Computer Science Ph.D. Candidate ylongqi@cs.cornell.edu Twitter: @ylongqi Funders: 1

Collabor Col aborator ators 2

Why Podc Wh dcast ast 3

Eme Emerging ng Int nterfa faces for Podcast Co Conte tent t Co Consump mpti tion 2

5

Wh What’ at’s s spec special al abo about t po podc dcasts asts (c (conten tent) t) … the architecture of the podcast is the precise antidote for the deep where now is shallow. It is flaws of the present. It is de from ads where now is completely vulnerable. It is a insulated fr thinking and refl chance for th flection ; it has an attention span an order of magnitude greater than the Tweet. It is an opportunity for serious (and playful) engagement. It is healthy eating for a brain-scape that now gorges on fast food. … --- Lawrence Lessig (Professor of Law at Harvard Law School) 6

Wh What’ at’s s spec special al abo about t po podc dcasts asts (c (conten tent) t) … It turns out, certain things humans can only do well if they do it sl slowly . Eating, cooking, reflecting, thinking, loving: These are the things we need to pace and pause … We should all spread the idea that every healthy mind spends time every week in slow thinking … --- Lawrence Lessig (Professor of Law at Harvard Law School) 7

Wh What’ at’s s spec special al abo about t po podc dcasts asts (u (user) ser) Past Future Fu (What you listened before) (What (W at you as aspire re to to liste ten in the future, user in in intentio ions and aspir iratio ions) 8

Wh What’ at’s s spec special al abo about t po podc dcasts asts (u (user) ser) People listened to episodes from subscribed channels (subscription-based consumption) 9

Compu Computati tation onal al Su Suppor pport f t for P or Podcasts odcasts articles Aa Aa posts music … rec. Past search 10

Compu Computati tation onal al Su Suppor pport f t for P or Podcasts odcasts articles Aa Aa posts Podcast music … rec. Past search 11

Compu Computati tation onal al Su Suppor pport f t for P or Podcasts odcasts articles Aa Aa posts Podcast music … rec. Past Podcast search 12

Age Agend nda More than Just Words (WSDM’19) Debias Offline Recommendation Evaluation (Recsys’18) Intention Informed Recommendations (Under Review) 13

Con Conten tent == t == Words ords Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 14

Po Podcast Content == Words (i (iTunes Podcas ast t dire recto tory) Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 15

Po Podcast Content > Words Conversational Paralinguistic Musical Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 16

Po Podcast Content > Words Conversational Paralinguistic Musical https://podcastfasttrack.com/podcast-editing/ Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 17

Our Goal: Mod Our Goal: Modeling eling Non Non-textu textual al Ch Char aracter acteristi stics of cs of P Podcasts odcasts feature representation Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 18

A Naïv A Naïve Solution Solution MFCC IS09 IS13 … Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 19

A Naïv A Naïve Solution Solution MFCC IS09 Expected to be sub-optimal IS13 … Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 20

Our ap Our approach: roach: Unsup Unsuper ervised vised Rep Representation Lear resentation Learning ning large unlabeled podcast corpus Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 21

Our ap Our approach: roach: Unsup Unsuper ervised vised Rep Representation Lear resentation Learning ning Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 22

Our ap Our approach: roach: Unsup Unsuper ervised vised Rep Representation Lear resentation Learning ning Fine-grained variations Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 23

Adversar Ad arial Le ial Lear arning ning-based based Podc dcast ast Represen epresentati tation (A (ALPR PR) vectors sampled from a uniform distribution Generator features (ALPR) Discriminator CE Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 24

Ad Adversar arial Le ial Lear arning ning-based based Podc dcast ast Represen epresentati tation (A (ALPR PR) Train the generator vectors sampled from a uniform distribution features (ALPR) Label=1 (real) Generator Discriminator CE Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 25

Adversar Ad arial Le ial Lear arning ning-based based Podc dcast ast Represen epresentati tation (A (ALPR PR) Train the discriminator and the classifier Label=1 (real) spectrograms of real podcast audio CE Discriminator Generator CE Label=0 (generated) Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 26

Adversar Ad arial Le ial Lear arning ning-based based Podc dcast ast Represen epresentati tation (A (ALPR PR) The generator 1 64 128 128 x 256 512 64 32 16 8 32 64 512 128 256 z fully deconv, 5x5 deconv, 5x5 deconv, 5x5 deconv, 5x5 connected stride 2 stride 2 stride 2 stride 2 Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 27

Ad Adversar arial Le ial Lear arning ning-based based Podc dcast ast Represen epresentati tation (A (ALPR PR) The discriminator conv, 5x5 conv, 5x5 conv, 5x5 conv, 5x5 stride 2 stride 2 stride 2 stride 2 x 512 64 32 16 8 32 128 64 256 512 global average pooling 256 128 128 64 fully connected 1 D(x) Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 28

Ad Adversar arial Le ial Lear arning ning-based based Podc dcast ast Represen epresentati tation (A (ALPR PR) Corpus: 728 episodes 88, 88,728 ( 18, 433 channels) 18,433 Training: Evaluation: 42,370 42, 370 episodes 46,358 46, 358 episodes Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 29

Ev Evalua luations ions Attr Attrib ibute utes Clas Classif ification ication (binar inary) y) Calm vs. Energetic Humorous vs. Serious Po Popularity prediction (binary) Top channels on iTunes vs. Others Longqi Yang, Yu Wang, Drew Dunne, Michael Sobolev, Mor Naaman, and Deborah Estrin. More than just words: Modeling non-textual characteristics of podcasts. In 12th ACM International Conference on Web Search and Data Mining (WSDM), 2019. 30