LY
Uploaded byLongqi Yang
PPTX, PDF641 views

Immersive Recommendation Workshop, NYC Media Lab'17

The document discusses advancements in recommendation systems, focusing on deep user and content modeling for personalized experiences. It highlights the transition from traditional item-based filtering to more complex models that consider diverse user interactions and contexts. The work also introduces 'yum-me', an interactive meal recommendation system aimed at promoting healthy eating based on personal dietary preferences.

Engineering◦
Related topics:
Deep Learning•

Embed presentation

Downloaded 28 times
Immersive Recommendation Deep User and Content Modeling for Personalization Longqi Yang, Ph.D. student Connected Experiences Lab, Small Data Lab Cornell Tech
Collaborators Faculty Interns Industry Collaborators Ph.D. students PostDocs
What we will be talking about today Recommendation Systems: Past, Present and Future Immersive Recommendation. OpenRec.
Recommendation Systems Research
The Matrix From item-based filtering to Netflix Challenge
… … … … … … … … … Item-based filtering (WWW 2001) ?4 1 1 4 5 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001. 1 5 1
… … … … … … … … … Item-based filtering (WWW 2001) ?4 1 1 1 5 4 5 4 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
… … … … … … … … … Item-based filtering (WWW 2001) ?4 1 1 1 5 4 5 4 0.9 0.1 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
… … … … … … … … … Item-based filtering (WWW 2001) ?4 1 1 1 5 4 5 4 0.9 0.1 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
… … … … … … … … … Item-based filtering (WWW 2001) 3.74 1 1 1 5 4 5 4 0.9 0.1 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
Early Adoption by Amazon.com
Netflix Challenge (Prize) We’re quite curious, really. To the tune of one million dollars… … To help customers find movies, we’ve developed our world-class movie recommendation system: Cinematch. Its job is to predict whether someone will enjoy a movie based on how much they liked or disliked other movies … … We provide you with a lot of anonymous rating data, and a prediction accuracy bar that is 10% better than what Cinematch can do on the same training data set… If you develop a system that we judge most beats that bar on the qualifying test set we provide, you get serious money and the bragging rights …
Netflix Challenge (Prize) Cinematch score - RMSE = 0.9525 2007 Progress Prize - RMSE = 0.8723 8.42% 2008 Progress Prize - RMSE = 0.8627 9.27% 2009 Grand Prize - RMSE = 0.8567 10.06% (AT&T Research)
Rating-based recommendations
Beyond Matrix Implicit feedback Deep Content Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
Beyond Matrix Implicit feedback Deep Content Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
Click-through Upvote/Like View/Watch/Visit Listen/Play Implicit Feedback
Main Challenge of Implicit Feedback Only “positive signal” is observed Does “rating estimation” still work?
… … … … … … … … … ?1 1 1 1 1 1 1 1 Main Challenge of Implicit Feedback
… … … … … … … … … 11 1 1 1 1 1 1 1 Main Challenge of Implicit Feedback
… … … … … … … … … 11 1 1 1 1 1 1 1 No matter what this item is! Main Challenge of Implicit Feedback
Feature Learning Framework for Implicit Feedback [0.1 -0.2 0.35 … 0.15]𝒖 𝟏 = 𝒗 𝟏 = [-0.05 0.5 0.1 … -0.3] 𝒖 𝟐 𝒖 𝑵 … 𝒗 𝟐 𝒗 𝑴 … Optimization
One Example – Bayesian Personalized Ranking (BPR) 𝒖𝒊, 𝒗 𝒑, 𝒗 𝒏For all m𝑎𝑥 ln 𝜎 𝒖𝒊 ∙ 𝒗 𝒑 − 𝒖𝒊 ∙ 𝒗 𝒏 Rendle, Steffen, et al. "BPR: Bayesian personalized ranking from implicit feedback." Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence. AUAI Press, 2009. user i An item that the user “click” An item that the user does not “click”
Algorithms (Incomprehensive List) • Weighted Regularized Matrix Factorization (WRMF) • Probabilistic Matrix Factorization (PMF) “Shallow” Models: • Weighted Approximately Ranked Pairwise Loss (WARP) “Deep” Models: Hsieh, Cheng-Kang, et al. "Collaborative metric learning." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017.• Collaborative Metric Learning (CML) He, Xiangnan, et al. "Neural collaborative filtering." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017.• Neural Collaborative Filtering Mnih, Andriy, and Ruslan R. Salakhutdinov. "Probabilistic matrix factorization." Advances in neural information processing systems. 2008. • Wide and Deep Learning for Recommender Systems Hu, Yifan, Yehuda Koren, and Chris Volinsky. "Collaborative filtering for implicit feedback datasets." Data Mining, 2008. ICDM'08. Eighth IEEE International Conference on. Ieee, 2008. Cheng, Heng-Tze, et al. "Wide & deep learning for recommender systems." Proceedings of the 1st Workshop on Deep Learning for Recommender Systems. ACM, 2016. Weston, Jason, Samy Bengio, and Nicolas Usunier. "Wsabie: Scaling up to large vocabulary image annotation." IJCAI. Vol. 11. 2011.
Beyond Matrix Implicit feedback Deep Content Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
Pure Collaborative Filtering is cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! - Articles Is it appropriate to recommend these two articles together?
Pure Collaborative Filtering is cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! - Food Random (Most healthy) Trattner, Christoph, and David Elsweiler. "Investigating the healthiness of internet- sourced recipes: implications for meal planning and recommender systems." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017
Pure Collaborative Filtering is cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! - Food “Users in general tend to interact most often with the least healthy recipes. Recommender algorithms tend to score popular items highly and thus on average promote unhealthy items.” Trattner, Christoph, and David Elsweiler. "Investigating the healthiness of internet- sourced recipes: implications for meal planning and recommender systems." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017
Pure Collaborative Filtering is cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! – Cold Start A new fashion cloth A new online course A new job post Recommendations without user feedback
Deep Content Modeling for Recommendations (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization
Beyond Matrix Implicit feedback Deep Content Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
Are ratings/clicks/views enough for recommendations? Context matters! – Music Recommendation Schedl, Markus, et al. "Music recommender systems." Recommender Systems Handbook. Springer US, 2015. 453-492. Schedl, Markus, Peter Knees, and Fabien Gouyon. "New Paths in Music Recommender Systems Research." Proceedings of the Eleventh ACM Conference on Recommender Systems. ACM, 2017. time location weather Environmental Context Individual Context emotion activity social context Schedl, Markus, Georg Breitschopf, and Bogdan Ionescu. "Mobile Music Genius: Reggae at the Beach, Metal on a Friday Night?." Proceedings of International Conference on Multimedia Retrieval. ACM, 2014.
Recommendations are not always “a list”: Rich modality Sun, Yu, et al. "Contextual intent tracking for personal assistants." Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016. Kang, Jie, et al. "Understanding How People Use Natural Language to Ask for Recommendations." Proceedings of the Eleventh ACM Conference on Recommender Systems. ACM, 2017.
Rich Context and Modality (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization (C/R/Res/Adversarial/Rei nforcement) NN (C/R/Res/Adversarial/Rei nforcement) NN
Beyond Matrix Implicit feedback Deep Content Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
Accuracy Popularity & Similarity
Diversity – Filtering bubble and echo chamber
Fairness – Long tail and Minority # views (attention) popular unpopular
Fairness – Long tail and Minority Recommender system better worse Yao, Sirui, and Bert Huang. "Beyond Parity: Fairness Objectives for Collaborative Filtering." arXiv preprint arXiv:1705.08804 (2017).
Incorporating diversity and fairness into recommendations (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization (C/R/Res/Adversarial/Rei nforcement) NN (C/R/Res/Adversarial/Rei nforcement) NN Penalize homogeneous and unfair recommendations
Immersive Recommendation
Immersive Recommendation Deep understanding of users’ diverse digital traces Deep modeling of heterogeneous contents + = News Events Food Art Spoken word
Immersive Recommendation Deep understanding of users’ diverse digital traces Deep modeling of heterogeneous contents + = News Events Food Art Spoken word
Yum-me Bringing healthiness into the recommendation of food Yang, Longqi, et al. "Yum-Me: A Personalized Nutrient-Based Meal Recommender System." ACM Transactions on Information Systems (TOIS) 36.1 (2017): 7.
*Number of Americans Living with Diet-and Inactivity-Related Diseases Obesity HBP Diabetes 113M 50M 15M Critical Issue of Food
The problem is not awareness, but adherence How can we (efficiently) find meals that are healthy but also cater to people’s tastes? - Bringing the notion of healthiness into recommendations!
Yum-me: An interactive healthy meal recommendation system Take a look at the food below and tap all that look delicious to you. http:// http:// Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste 2iters + 13iters 2iters + 13iters 2iters + 13iters Browser Mobile Wearable Personal Dietary Profile (Food Preferences) … … Healthy meal recommendations based on dietary restrictions Re-ranking Personalized healthy meal recommendations …... …... Phase I Phase II Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste http:// Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian⌾ Vegan ⌾ Kosher ⌾ Halal Identify your health goals. ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase Calories Protein Fat + Survey Choose the closest diet type to you. Identify your health goals. ⌾Reduce ⌾ Maintain Calories ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal + + Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal
Yum-me: An interactive healthy meal recommendation system Take a look at the food below and tap all that look delicious to you. http:// http:// Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste 2iters + 13iters 2iters + 13iters 2iters + 13iters Browser Mobile Wearable Personal Dietary Profile (Food Preferences) … … Healthy meal recommendations based on dietary restrictions Re-ranking Personalized healthy meal recommendations …... …... Phase I Phase II Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste http:// Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian⌾ Vegan ⌾ Kosher ⌾ Halal Identify your health goals. ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase Calories Protein Fat + Survey Choose the closest diet type to you. Identify your health goals. ⌾Reduce ⌾ Maintain Calories ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal + + Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal
Interactive Learning Process propagate
Interactive Learning Process selection
Similarity between Food Items 0.9151 0.6471 0.9652 1.3484 1.3410 1.3484 1.1476 Siamese Network
Similarity between Food Items (a) No restrictions (b) Vegetarian
Yum-me: An interactive healthy meal recommendation system Take a look at the food below and tap all that look delicious to you. http:// http:// Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste 2iters + 13iters 2iters + 13iters 2iters + 13iters Browser Mobile Wearable Personal Dietary Profile (Food Preferences) … … Healthy meal recommendations based on dietary restrictions Re-ranking Personalized healthy meal recommendations …... …... Phase I Phase II Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste http:// Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian⌾ Vegan ⌾ Kosher ⌾ Halal Identify your health goals. ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase Calories Protein Fat + Survey Choose the closest diet type to you. Identify your health goals. ⌾Reduce ⌾ Maintain Calories ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal + + Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal
Re-ranking healthy unhealthy Re-ranked by an user’s preference Top N recommendations
User study Step 1. Users identify their diet types and health goals. Step 2. Users use visual interace to express their fine-grained food preferences. Step 3. Users identify each of recommended meals as either Yummy or No way. (The order of the items is randomized) Top 500 healthy items that meet users’ diet types and health goals. Select top 10 items ranked by user’s fine-grained dietary preference. Randomly select 10 food items from 500 healthy meal pool. …... …... …...…...
Acceptance Rate 51% 72.5% Traditional Approach Yum-me
User study Goal: reduce calories (25 users) Goal: maintain calories (21 users) Goal: maintain protein (36 users) Goal: increase protein (12 users) Goal: reduce fat (17 users) Goal: increase calories (2 users) Goal: maintain fat (30 users) users’ 20 favorite meals meals recommended by Yum-me and accepted by users. Averageamountofnutrients perserving(normalized) Averageamountofnutrients perserving(normalized) Averageamountofnutrients perserving(normalized)
Creative Content Recommendation Bringing unstructured command traces into the recommendation of art Yang, Longqi, et al. "Personalizing Software and Web Services by Integrating Unstructured Application Usage Traces." Proceedings of the 26th International Conference on World Wide Web Companion. International World Wide Web Conferences Steering Committee, 2017.
Cold-start creative content recommendation Day to day work activities (Commands performed in professional design software)
… FC FC FC FC Two-step recommendation algorithm
utilization-to-vector (util2vec) action action action action action Representation Learning Paradigm
utilization-to-vector (util2vec) - sliding window sliding window
utilization-to-vector (util2vec) - sliding window sliding window
utilization-to-vector (util2vec) - inside each window prediction target 2n+1 actions (n=4) predictor inputs
utilization-to-vector (util2vec) - inside each window prediction target 2n+1 actions (n=4) predictor inputs
Concatenation/Average Softmax Predictor utilization-to-vector (util2vec) - predictor
Evaluation
OpenRec An open source and modular framework for extensible recommendation algorithms
Recap (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization (C/R/Res/Adversarial/Rei nforcement) NN (C/R/Res/Adversarial/Rei nforcement) NN
Real World Systems can only be more complex User Item Interaction Optimization … … … … … … User Item Interaction User Item Interaction
A Traditional (Monolithic) View of Recommender Systems … RS 1 RS 2 RS 3 RS n
OpenRec: Experimentation and innovation through Extension rating item text item textitem image rating user text item image item text rating rating user text user demogr item text item image integrator module extractor module interaction module R1 R2 R4R3
OpenRec: Architecture Module Extractor IntegratorInteraction BPR WARP PMF CML NeuMF … LF ResNet MLP LSTM FoodDist Concatenation Average Weighted sum …… Recommender News recommender system with users’ click history, Twitter posts and news topic modeling. Music recommender system with users’ listen history, lyrics and audio analysis … Utility Sampler Pairwise sampler Triplet sampler … Evaluator … AUC Recall@K
OpenRec: Examples input buff input holders user extractors item extractors interactions optimizerVisualCML Modules VBPR input buff input holders user extractors item extractors interactions optimizer reusable modules reusable functions build from scratch CML Interactionmodule … input buff input holders user extractors item extractors interactions optimizerVanillaBPR Modules reusable modules build from scratch LatentFactor BPR Extractor module Interactionmodule … …
Stay tuned, Late Fall: https://github.com/ylongqi/OpenRec OpenRec
Thank you! Longqi Yang Ph.D. Student, Computer Science, Cornell Tech, Cornell University Email: ylongqi@cs.cornell.edu Web: bit.ly/longqi Twitter: @ylongqi Connected Experiences Lab: http://cx.jacobs.cornell.edu/ Small Data Lab: http://smalldata.io/

More Related Content

PDF
Altman - Perfectly Anonymous Data is Perfectly Useless Data
byNational Information Standards Organization (NISO)
 
PPTX
Writing Analytics for Epistemic Features of Student Writing #icls2016 talk
bySimon Knight
 
PPTX
How to create a taxonomy for management buy-in
byMary Chitty
 
PPTX
2015 Kno.e.sis Center Annual Review
byArtificial Intelligence Institute at UofSC
 
PDF
Trust Management: A Tutorial
byArtificial Intelligence Institute at UofSC
 
PPTX
Panel: Our Scholarly Recognition System Doesn’t Still Work
byDaniel S. Katz
 
PDF
Expertise for the future: harnessing the power of digital technologies
byEFSA EU
 
PPTX
Epistemic networks for Epistemic Commitments
bySimon Knight
 
Altman - Perfectly Anonymous Data is Perfectly Useless Data
byNational Information Standards Organization (NISO)
 
Writing Analytics for Epistemic Features of Student Writing #icls2016 talk
bySimon Knight
 
How to create a taxonomy for management buy-in
byMary Chitty
 
2015 Kno.e.sis Center Annual Review
byArtificial Intelligence Institute at UofSC
 
Trust Management: A Tutorial
byArtificial Intelligence Institute at UofSC
 
Panel: Our Scholarly Recognition System Doesn’t Still Work
byDaniel S. Katz
 
Expertise for the future: harnessing the power of digital technologies
byEFSA EU
 
Epistemic networks for Epistemic Commitments
bySimon Knight
 

What's hot

PDF
Tell It Like It Seems: Challenges Identifying Requirements of a Learning He...
byDepartment of Learning Health Sciences, University of Michigan Medical School
 
PPTX
From metasearch to metaservices
bydswalker
 
PPTX
How to find out about the usability of your web site using a survey by @cjforms
byCaroline Jarrett
 
PDF
Mechanical Librarian
byAndre Vellino
 
PDF
Data Sets as Facilitator for new Products and Services for Universities
byHendrik Drachsler
 
PPTX
What's up at Kno.e.sis?
byAmit Sheth
 
PDF
Online Communities in Citizen Science
byAndrea Wiggins
 
KEY
Open Source, Open Science, & Citizen Science
byAndrea Wiggins
 
PPTX
Anonymous peer review062016 shared
byGjoa Andrichuk
 
PPTX
CHI2015 - Citizen Science || Zooniverse
byRamine Tinati
 
PDF
Free as in Puppies: Compensating for ICT Constraints in Citizen Science
byAndrea Wiggins
 
PPTX
Data, Responsibly: The Next Decade of Data Science
byUniversity of Washington
 
PPT
Meyer primr-sililcon flatirons expanded facebook slide deck
byMichelle N. Meyer
 
PPT
Kno.e.sis Review: late 2012 to mid 2013
byArtificial Intelligence Institute at UofSC
 
PDF
Learning Analytics vs Cognitive Automation
bySimon Buckingham Shum
 
PDF
Recommender system a-introduction
byzh3f
 
PPTX
Introduction and new trends in Recommender Systems
byPaolo Tomeo
 
PPTX
Knoesis Student Achievement
byArtificial Intelligence Institute at UofSC
 
PPTX
Crowdsourcing and Learning from Crowd Data (Tutorial @ PSB2015)
byAndrew Su
 
PPTX
Science Data, Responsibly
byUniversity of Washington
 
Tell It Like It Seems: Challenges Identifying Requirements of a Learning He...
byDepartment of Learning Health Sciences, University of Michigan Medical School
 
From metasearch to metaservices
bydswalker
 
How to find out about the usability of your web site using a survey by @cjforms
byCaroline Jarrett
 
Mechanical Librarian
byAndre Vellino
 
Data Sets as Facilitator for new Products and Services for Universities
byHendrik Drachsler
 
What's up at Kno.e.sis?
byAmit Sheth
 
Online Communities in Citizen Science
byAndrea Wiggins
 
Open Source, Open Science, & Citizen Science
byAndrea Wiggins
 
Anonymous peer review062016 shared
byGjoa Andrichuk
 
CHI2015 - Citizen Science || Zooniverse
byRamine Tinati
 
Free as in Puppies: Compensating for ICT Constraints in Citizen Science
byAndrea Wiggins
 
Data, Responsibly: The Next Decade of Data Science
byUniversity of Washington
 
Meyer primr-sililcon flatirons expanded facebook slide deck
byMichelle N. Meyer
 
Kno.e.sis Review: late 2012 to mid 2013
byArtificial Intelligence Institute at UofSC
 
Learning Analytics vs Cognitive Automation
bySimon Buckingham Shum
 
Recommender system a-introduction
byzh3f
 
Introduction and new trends in Recommender Systems
byPaolo Tomeo
 
Knoesis Student Achievement
byArtificial Intelligence Institute at UofSC
 
Crowdsourcing and Learning from Crowd Data (Tutorial @ PSB2015)
byAndrew Su
 
Science Data, Responsibly
byUniversity of Washington
 

Similar to Immersive Recommendation Workshop, NYC Media Lab'17

PPTX
Олександр Обєдніков “Рекомендательные системы”
byDakiry
 
PPTX
Reasesrty djhjan S - explanation required.pptx
byAnkitaVerma776806
 
PDF
Modern Perspectives on Recommender Systems and their Applications in Mendeley
byKris Jack
 
PPT
Introduction to recommendation system
byAravindharamanan S
 
PDF
Notes on Recommender Systems pdf 2nd module
byPRABHAKARK24
 
PDF
A flexible recommenndation system for Cable TV
byIntoTheMinds
 
PDF
A Flexible Recommendation System for Cable TV
byFrancisco Couto
 
PDF
Recommender.system.presentation.pjug.01.21.2014
byrpbrehm
 
PPT
Social Recommender Systems Tutorial - WWW 2011
byidoguy
 
PPTX
Rokach-GomaxSlides (1).pptx
byJadna Almeida
 
PPTX
Rokach-GomaxSlides.pptx
byJadna Almeida
 
PPT
Impersonal Recommendation system on top of Hadoop
byKostiantyn Kudriavtsev
 
PDF
Highlights from the 8th ACM Conference on Recommender Systems (RecSys 2014)
byDavid Zibriczky
 
PPTX
Collaborative Filtering Recommendation System
byMilind Gokhale
 
PDF
Modern Perspectives on Recommender Systems and their Applications in Mendeley
byMaya Hristakeva
 
PDF
Interactive Recommender Systems
byDenis Parra Santander
 
PDF
Past, present, and future of Recommender Systems: an industry perspective
byXavier Amatriain
 
PDF
Recent Trends in Personalization at Netflix
byFörderverein Technische Fakultät
 
PPT
Recommender systems
byTamer Rezk
 
PDF
Recommandation systems -
byYousef Fadila
 
Олександр Обєдніков “Рекомендательные системы”
byDakiry
 
Reasesrty djhjan S - explanation required.pptx
byAnkitaVerma776806
 
Modern Perspectives on Recommender Systems and their Applications in Mendeley
byKris Jack
 
Introduction to recommendation system
byAravindharamanan S
 
Notes on Recommender Systems pdf 2nd module
byPRABHAKARK24
 
A flexible recommenndation system for Cable TV
byIntoTheMinds
 
A Flexible Recommendation System for Cable TV
byFrancisco Couto
 
Recommender.system.presentation.pjug.01.21.2014
byrpbrehm
 
Social Recommender Systems Tutorial - WWW 2011
byidoguy
 
Rokach-GomaxSlides (1).pptx
byJadna Almeida
 
Rokach-GomaxSlides.pptx
byJadna Almeida
 
Impersonal Recommendation system on top of Hadoop
byKostiantyn Kudriavtsev
 
Highlights from the 8th ACM Conference on Recommender Systems (RecSys 2014)
byDavid Zibriczky
 
Collaborative Filtering Recommendation System
byMilind Gokhale
 
Modern Perspectives on Recommender Systems and their Applications in Mendeley
byMaya Hristakeva
 
Interactive Recommender Systems
byDenis Parra Santander
 
Past, present, and future of Recommender Systems: an industry perspective
byXavier Amatriain
 
Recent Trends in Personalization at Netflix
byFörderverein Technische Fakultät
 
Recommender systems
byTamer Rezk
 
Recommandation systems -
byYousef Fadila
 

Recently uploaded

PDF
Fluid Mechanics, Heat Transfer, and Mass Transfer_ Chemical Engineering Pract...
bynikptl09
 
PDF
0.39 Inch Micro-OLED Display 1024×768 XGA Panel Silicon OLED
bySyluxdisplay
 
PPTX
Basin Design Service, LLC Introduction Presentation
byHubie Fix
 
PPTX
SketchUp Pro 2026 – Advanced 3D Modeling Software
byHoka Moka
 
PDF
Manual de instalacion de notifire NFS320
byMarcoPolitoPerez
 
PDF
Model QP 2025 scheme Q &A- Module 1 and 2.pdf
bySURESHA V
 
PDF
Module 4 python programming-1BPLCK105B-2025 by Dr.SV.pdf
bySURESHA V
 
PPTX
Energy_Conservation_in_Cement_Factories.pptx
bysibirajmurugan8
 
PDF
engineering management chapter 5 ppt presentation
byericaangelatoledo06
 
PPTX
Week 1.pptx dynamics particle kinetics introduction
byfmsalih06
 
PPTX
Designing Work for Humans, Not Machines: Human-Centered Maintenance Excellence
byMaintWiz Technologies Private Limited
 
PDF
Python programming basics Unit-1 for R25 regulation
byssuser492e7f
 
PPTX
Data Link Layer Unit II Computer Networks.pptx
bylosstwolf17
 
PPTX
Unit 1_Importance of modelling(Object oriented concepts).pptx
byVidyaranichougule
 
PPTX
Integrating Azure Event Services_ Event Grid, Event Hubs, and Service Bus.pptx
byADITYASHARMA423788
 
PDF
BOQ LESSON 2-QUANTITY TAKE-OFF & RATE BUILD-UP.pdf
byGabrielTambwe1
 
PDF
Chris Elwell Woburn - An Experienced IT Executive
byChris Elwell Woburn
 
PDF
Lifting Operations Safety and Risk Control
bymanju774
 
PPTX
We-Optimized-Everything-Except-Decision-Quality-Maintenance-MaintWiz.pptx
byMaintWiz Technologies Private Limited
 
PDF
comprehensive analysis of cross-chain bridges in the DeFi - Garima Singh
byGarima Singh
 
Fluid Mechanics, Heat Transfer, and Mass Transfer_ Chemical Engineering Pract...
bynikptl09
 
0.39 Inch Micro-OLED Display 1024×768 XGA Panel Silicon OLED
bySyluxdisplay
 
Basin Design Service, LLC Introduction Presentation
byHubie Fix
 
SketchUp Pro 2026 – Advanced 3D Modeling Software
byHoka Moka
 
Manual de instalacion de notifire NFS320
byMarcoPolitoPerez
 
Model QP 2025 scheme Q &A- Module 1 and 2.pdf
bySURESHA V
 
Module 4 python programming-1BPLCK105B-2025 by Dr.SV.pdf
bySURESHA V
 
Energy_Conservation_in_Cement_Factories.pptx
bysibirajmurugan8
 
engineering management chapter 5 ppt presentation
byericaangelatoledo06
 
Week 1.pptx dynamics particle kinetics introduction
byfmsalih06
 
Designing Work for Humans, Not Machines: Human-Centered Maintenance Excellence
byMaintWiz Technologies Private Limited
 
Python programming basics Unit-1 for R25 regulation
byssuser492e7f
 
Data Link Layer Unit II Computer Networks.pptx
bylosstwolf17
 
Unit 1_Importance of modelling(Object oriented concepts).pptx
byVidyaranichougule
 
Integrating Azure Event Services_ Event Grid, Event Hubs, and Service Bus.pptx
byADITYASHARMA423788
 
BOQ LESSON 2-QUANTITY TAKE-OFF & RATE BUILD-UP.pdf
byGabrielTambwe1
 
Chris Elwell Woburn - An Experienced IT Executive
byChris Elwell Woburn
 
Lifting Operations Safety and Risk Control
bymanju774
 
We-Optimized-Everything-Except-Decision-Quality-Maintenance-MaintWiz.pptx
byMaintWiz Technologies Private Limited
 
comprehensive analysis of cross-chain bridges in the DeFi - Garima Singh
byGarima Singh
 

Immersive Recommendation Workshop, NYC Media Lab'17

  • 1.
    Immersive Recommendation Deep Userand Content Modeling for Personalization Longqi Yang, Ph.D. student Connected Experiences Lab, Small Data Lab Cornell Tech
  • 2.
  • 3.
    What we willbe talking about today Recommendation Systems: Past, Present and Future Immersive Recommendation. OpenRec.
  • 4.
  • 5.
    The Matrix From item-basedfiltering to Netflix Challenge
  • 6.
    … … … …… … … … … Item-based filtering (WWW 2001) ?4 1 1 4 5 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001. 1 5 1
  • 7.
    … … … …… … … … … Item-based filtering (WWW 2001) ?4 1 1 1 5 4 5 4 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
  • 8.
    … … … …… … … … … Item-based filtering (WWW 2001) ?4 1 1 1 5 4 5 4 0.9 0.1 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
  • 9.
    … … … …… … … … … Item-based filtering (WWW 2001) ?4 1 1 1 5 4 5 4 0.9 0.1 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
  • 10.
    … … … …… … … … … Item-based filtering (WWW 2001) 3.74 1 1 1 5 4 5 4 0.9 0.1 Sarwar, Badrul, et al. "Item-based collaborative filtering recommendation algorithms." Proceedings of the 10th international conference on World Wide Web. ACM, 2001.
  • 11.
  • 12.
    Netflix Challenge (Prize) We’requite curious, really. To the tune of one million dollars… … To help customers find movies, we’ve developed our world-class movie recommendation system: Cinematch. Its job is to predict whether someone will enjoy a movie based on how much they liked or disliked other movies … … We provide you with a lot of anonymous rating data, and a prediction accuracy bar that is 10% better than what Cinematch can do on the same training data set… If you develop a system that we judge most beats that bar on the qualifying test set we provide, you get serious money and the bragging rights …
  • 13.
    Netflix Challenge (Prize) Cinematchscore - RMSE = 0.9525 2007 Progress Prize - RMSE = 0.8723 8.42% 2008 Progress Prize - RMSE = 0.8627 9.27% 2009 Grand Prize - RMSE = 0.8567 10.06% (AT&T Research)
  • 14.
  • 15.
    Beyond Matrix Implicit feedback DeepContent Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
  • 16.
    Beyond Matrix Implicit feedback DeepContent Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
  • 17.
  • 18.
    Main Challenge ofImplicit Feedback Only “positive signal” is observed Does “rating estimation” still work?
  • 19.
    … … … …… … … … … ?1 1 1 1 1 1 1 1 Main Challenge of Implicit Feedback
  • 20.
    … … … …… … … … … 11 1 1 1 1 1 1 1 Main Challenge of Implicit Feedback
  • 21.
    … … … …… … … … … 11 1 1 1 1 1 1 1 No matter what this item is! Main Challenge of Implicit Feedback
  • 22.
    Feature Learning Frameworkfor Implicit Feedback [0.1 -0.2 0.35 … 0.15]𝒖 𝟏 = 𝒗 𝟏 = [-0.05 0.5 0.1 … -0.3] 𝒖 𝟐 𝒖 𝑵 … 𝒗 𝟐 𝒗 𝑴 … Optimization
  • 23.
    One Example –Bayesian Personalized Ranking (BPR) 𝒖𝒊, 𝒗 𝒑, 𝒗 𝒏For all m𝑎𝑥 ln 𝜎 𝒖𝒊 ∙ 𝒗 𝒑 − 𝒖𝒊 ∙ 𝒗 𝒏 Rendle, Steffen, et al. "BPR: Bayesian personalized ranking from implicit feedback." Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence. AUAI Press, 2009. user i An item that the user “click” An item that the user does not “click”
  • 24.
    Algorithms (Incomprehensive List) •Weighted Regularized Matrix Factorization (WRMF) • Probabilistic Matrix Factorization (PMF) “Shallow” Models: • Weighted Approximately Ranked Pairwise Loss (WARP) “Deep” Models: Hsieh, Cheng-Kang, et al. "Collaborative metric learning." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017.• Collaborative Metric Learning (CML) He, Xiangnan, et al. "Neural collaborative filtering." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017.• Neural Collaborative Filtering Mnih, Andriy, and Ruslan R. Salakhutdinov. "Probabilistic matrix factorization." Advances in neural information processing systems. 2008. • Wide and Deep Learning for Recommender Systems Hu, Yifan, Yehuda Koren, and Chris Volinsky. "Collaborative filtering for implicit feedback datasets." Data Mining, 2008. ICDM'08. Eighth IEEE International Conference on. Ieee, 2008. Cheng, Heng-Tze, et al. "Wide & deep learning for recommender systems." Proceedings of the 1st Workshop on Deep Learning for Recommender Systems. ACM, 2016. Weston, Jason, Samy Bengio, and Nicolas Usunier. "Wsabie: Scaling up to large vocabulary image annotation." IJCAI. Vol. 11. 2011.
  • 25.
    Beyond Matrix Implicit feedback DeepContent Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
  • 26.
    Pure Collaborative Filteringis cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! - Articles Is it appropriate to recommend these two articles together?
  • 27.
    Pure Collaborative Filteringis cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! - Food Random (Most healthy) Trattner, Christoph, and David Elsweiler. "Investigating the healthiness of internet- sourced recipes: implications for meal planning and recommender systems." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017
  • 28.
    Pure Collaborative Filteringis cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! - Food “Users in general tend to interact most often with the least healthy recipes. Recommender algorithms tend to score popular items highly and thus on average promote unhealthy items.” Trattner, Christoph, and David Elsweiler. "Investigating the healthiness of internet- sourced recipes: implications for meal planning and recommender systems." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017
  • 29.
    Pure Collaborative Filteringis cool (and maybe accurate), but real world recommendations are far more complex than “likes” Understanding the contents really matters! – Cold Start A new fashion cloth A new online course A new job post Recommendations without user feedback
  • 30.
    Deep Content Modelingfor Recommendations (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization
  • 31.
    Beyond Matrix Implicit feedback DeepContent Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
  • 32.
    Are ratings/clicks/views enoughfor recommendations? Context matters! – Music Recommendation Schedl, Markus, et al. "Music recommender systems." Recommender Systems Handbook. Springer US, 2015. 453-492. Schedl, Markus, Peter Knees, and Fabien Gouyon. "New Paths in Music Recommender Systems Research." Proceedings of the Eleventh ACM Conference on Recommender Systems. ACM, 2017. time location weather Environmental Context Individual Context emotion activity social context Schedl, Markus, Georg Breitschopf, and Bogdan Ionescu. "Mobile Music Genius: Reggae at the Beach, Metal on a Friday Night?." Proceedings of International Conference on Multimedia Retrieval. ACM, 2014.
  • 33.
    Recommendations are notalways “a list”: Rich modality Sun, Yu, et al. "Contextual intent tracking for personal assistants." Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016. Kang, Jie, et al. "Understanding How People Use Natural Language to Ask for Recommendations." Proceedings of the Eleventh ACM Conference on Recommender Systems. ACM, 2017.
  • 34.
    Rich Context andModality (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization (C/R/Res/Adversarial/Rei nforcement) NN (C/R/Res/Adversarial/Rei nforcement) NN
  • 35.
    Beyond Matrix Implicit feedback DeepContent Modeling Beyond “black-box items” Beyond Accuracy Diversity and Fairness Rich Context and Modality Learning preference from auxiliary channels
  • 36.
  • 37.
    Diversity – Filteringbubble and echo chamber
  • 38.
    Fairness – Longtail and Minority # views (attention) popular unpopular
  • 39.
    Fairness – Longtail and Minority Recommender system better worse Yao, Sirui, and Bert Huang. "Beyond Parity: Fairness Objectives for Collaborative Filtering." arXiv preprint arXiv:1705.08804 (2017).
  • 40.
    Incorporating diversity andfairness into recommendations (C/R/Res/Adversarial/Rei nforcement) NN User Item Interaction Optimization (C/R/Res/Adversarial/Rei nforcement) NN (C/R/Res/Adversarial/Rei nforcement) NN Penalize homogeneous and unfair recommendations
  • 41.
  • 42.
    Immersive Recommendation Deep understandingof users’ diverse digital traces Deep modeling of heterogeneous contents + = News Events Food Art Spoken word
  • 43.
    Immersive Recommendation Deep understandingof users’ diverse digital traces Deep modeling of heterogeneous contents + = News Events Food Art Spoken word
  • 44.
    Yum-me Bringing healthiness intothe recommendation of food Yang, Longqi, et al. "Yum-Me: A Personalized Nutrient-Based Meal Recommender System." ACM Transactions on Information Systems (TOIS) 36.1 (2017): 7.
  • 45.
    *Number of AmericansLiving with Diet-and Inactivity-Related Diseases Obesity HBP Diabetes 113M 50M 15M Critical Issue of Food
  • 46.
    The problem isnot awareness, but adherence How can we (efficiently) find meals that are healthy but also cater to people’s tastes? - Bringing the notion of healthiness into recommendations!
  • 47.
    Yum-me: An interactivehealthy meal recommendation system Take a look at the food below and tap all that look delicious to you. http:// http:// Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste 2iters + 13iters 2iters + 13iters 2iters + 13iters Browser Mobile Wearable Personal Dietary Profile (Food Preferences) … … Healthy meal recommendations based on dietary restrictions Re-ranking Personalized healthy meal recommendations …... …... Phase I Phase II Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste http:// Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian⌾ Vegan ⌾ Kosher ⌾ Halal Identify your health goals. ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase Calories Protein Fat + Survey Choose the closest diet type to you. Identify your health goals. ⌾Reduce ⌾ Maintain Calories ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal + + Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal
  • 48.
    Yum-me: An interactivehealthy meal recommendation system Take a look at the food below and tap all that look delicious to you. http:// http:// Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste 2iters + 13iters 2iters + 13iters 2iters + 13iters Browser Mobile Wearable Personal Dietary Profile (Food Preferences) … … Healthy meal recommendations based on dietary restrictions Re-ranking Personalized healthy meal recommendations …... …... Phase I Phase II Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste http:// Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian⌾ Vegan ⌾ Kosher ⌾ Halal Identify your health goals. ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase Calories Protein Fat + Survey Choose the closest diet type to you. Identify your health goals. ⌾Reduce ⌾ Maintain Calories ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal + + Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal
  • 49.
  • 50.
  • 51.
    Similarity between FoodItems 0.9151 0.6471 0.9652 1.3484 1.3410 1.3484 1.1476 Siamese Network
  • 52.
    Similarity between FoodItems (a) No restrictions (b) Vegetarian
  • 53.
    Yum-me: An interactivehealthy meal recommendation system Take a look at the food below and tap all that look delicious to you. http:// http:// Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste 2iters + 13iters 2iters + 13iters 2iters + 13iters Browser Mobile Wearable Personal Dietary Profile (Food Preferences) … … Healthy meal recommendations based on dietary restrictions Re-ranking Personalized healthy meal recommendations …... …... Phase I Phase II Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste Take a look at the food below and tap all that look delicious to you. Compare the food pair below and tap on whichever looks delicious to you. Press on Yuck if neither of them fits to your taste http:// Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian⌾ Vegan ⌾ Kosher ⌾ Halal Identify your health goals. ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase ⌾Reduce ⌾ Maintain ⌾ Increase Calories Protein Fat + Survey Choose the closest diet type to you. Identify your health goals. ⌾Reduce ⌾ Maintain Calories ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal + + Choose the closest diet type to you. ⌾No restrictions ⌾ Vegetarian ⌾ Vegan ⌾ Kosher ⌾ Halal
  • 54.
    Re-ranking healthy unhealthy Re-ranked byan user’s preference Top N recommendations
  • 55.
    User study Step 1.Users identify their diet types and health goals. Step 2. Users use visual interace to express their fine-grained food preferences. Step 3. Users identify each of recommended meals as either Yummy or No way. (The order of the items is randomized) Top 500 healthy items that meet users’ diet types and health goals. Select top 10 items ranked by user’s fine-grained dietary preference. Randomly select 10 food items from 500 healthy meal pool. …... …... …...…...
  • 56.
  • 57.
    User study Goal: reducecalories (25 users) Goal: maintain calories (21 users) Goal: maintain protein (36 users) Goal: increase protein (12 users) Goal: reduce fat (17 users) Goal: increase calories (2 users) Goal: maintain fat (30 users) users’ 20 favorite meals meals recommended by Yum-me and accepted by users. Averageamountofnutrients perserving(normalized) Averageamountofnutrients perserving(normalized) Averageamountofnutrients perserving(normalized)
  • 58.
    Creative Content Recommendation Bringingunstructured command traces into the recommendation of art Yang, Longqi, et al. "Personalizing Software and Web Services by Integrating Unstructured Application Usage Traces." Proceedings of the 26th International Conference on World Wide Web Companion. International World Wide Web Conferences Steering Committee, 2017.
  • 59.
    Cold-start creative contentrecommendation Day to day work activities (Commands performed in professional design software)
  • 60.
  • 61.
    utilization-to-vector (util2vec) action actionaction action action Representation Learning Paradigm
  • 62.
    utilization-to-vector (util2vec) -sliding window sliding window
  • 63.
    utilization-to-vector (util2vec) -sliding window sliding window
  • 64.
    utilization-to-vector (util2vec) -inside each window prediction target 2n+1 actions (n=4) predictor inputs
  • 65.
    utilization-to-vector (util2vec) -inside each window prediction target 2n+1 actions (n=4) predictor inputs
  • 66.
  • 67.
  • 68.
    OpenRec An open sourceand modular framework for extensible recommendation algorithms
  • 69.
    Recap (C/R/Res/Adversarial/Rei nforcement) NN User ItemInteraction Optimization (C/R/Res/Adversarial/Rei nforcement) NN (C/R/Res/Adversarial/Rei nforcement) NN
  • 70.
    Real World Systemscan only be more complex User Item Interaction Optimization … … … … … … User Item Interaction User Item Interaction
  • 71.
    A Traditional (Monolithic)View of Recommender Systems … RS 1 RS 2 RS 3 RS n
  • 72.
    OpenRec: Experimentation andinnovation through Extension rating item text item textitem image rating user text item image item text rating rating user text user demogr item text item image integrator module extractor module interaction module R1 R2 R4R3
  • 73.
    OpenRec: Architecture Module Extractor IntegratorInteraction BPRWARP PMF CML NeuMF … LF ResNet MLP LSTM FoodDist Concatenation Average Weighted sum …… Recommender News recommender system with users’ click history, Twitter posts and news topic modeling. Music recommender system with users’ listen history, lyrics and audio analysis … Utility Sampler Pairwise sampler Triplet sampler … Evaluator … AUC Recall@K
  • 74.
    OpenRec: Examples input buffinput holders user extractors item extractors interactions optimizerVisualCML Modules VBPR input buff input holders user extractors item extractors interactions optimizer reusable modules reusable functions build from scratch CML Interactionmodule … input buff input holders user extractors item extractors interactions optimizerVanillaBPR Modules reusable modules build from scratch LatentFactor BPR Extractor module Interactionmodule … …
  • 75.
    Stay tuned, LateFall: https://github.com/ylongqi/OpenRec OpenRec
  • 76.
    Thank you! Longqi Yang Ph.D.Student, Computer Science, Cornell Tech, Cornell University Email: ylongqi@cs.cornell.edu Web: bit.ly/longqi Twitter: @ylongqi Connected Experiences Lab: http://cx.jacobs.cornell.edu/ Small Data Lab: http://smalldata.io/