session 01-jianling wang-user recommendation in content...
TRANSCRIPT
![Page 1: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/1.jpg)
User Recommendation in Content Curation Platforms
Jianling Wang, Ziwei Zhu and James Caverlee
![Page 2: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/2.jpg)
Content creators generate new digital artifacts such as tweets, blog posts, or photos.
Content Creation vs Curation
![Page 3: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/3.jpg)
Music Streaming platforms allow users to create and share playlists.
Mega Hit Mix
Jogging!
Mood Booster
Playlists
Content Creation vs Curation
![Page 4: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/4.jpg)
John rated it John added to Bio
Like Comment Preview Book
John Smith Following
John added to want to read
Like Comment Preview Book Like Comment Preview Book
Content Creation vs CurationGoodreads provide a platform for users to curate interesting books via tagging, ratings and reviews.
![Page 5: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/5.jpg)
In Content Curation Platforms, users acting as curators, collect and organize existing content via reviews, pins, boards, ratings and other actions.
Our Goal: Recommend Curators
![Page 6: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/6.jpg)
Our Goal: Recommend CuratorsCompared with:
• Item-level recommendation, e.g., recommend music tracks
There are many new items or items with little feedback.
• Curation-level recommendation, e.g., recommend playlists
Curations (e.g. pin boards, playlists) are frequently updated.
![Page 7: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/7.jpg)
• Curators can provide a human-powered overlay that can link seemingly un-related items (e.g., a collection of songs that are thematically related though from different genres).
Why Recommend Curators?
For study 📚
Jogging!
Mood Booster
Playlists
![Page 8: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/8.jpg)
• By receiving updates from whom they follow, users can be exposed to interesting items and curation decisions.
Get Update
Follow
Why Recommend Curators?
RateTag Listen
Item
User
![Page 9: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/9.jpg)
Our SettingWe can collect:
• User-curator following relationships
• Implicit feedback on items
TagRead
Highlight
TagRead
Highlight
Follow0100
10
…
1001
01
…
Feedback Vector on Items
Feedback Vector on Curators
![Page 10: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/10.jpg)
ChallengeHow to model these two aspects - curator preferences and item preferences - in a unified model?
TagRead
Highlight
TagRead
Highlight
Follow0100
10
…
1001
01
…
Feedback Vector on Items
Feedback Vector on Curators
![Page 11: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/11.jpg)
The GoalWe are motivated to develop a new model for Curator Recommendation that leverages the linkage between user-curator following relationships and the items they are interested in.
![Page 12: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/12.jpg)
The Joint TasksUltimately, the model aims to provide users with recommendation on:
• who to follow (the primary task)
• interesting items (the supplementary task)
![Page 13: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/13.jpg)
CuRe - Curator RecommendationThree components:
• Learning Curator & Item Preferences
• Fusing Latent Representations
• Personalized via Attention
![Page 14: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/14.jpg)
Uncover the Preferences
Follow
0100
10
… hCV
0100
10
…
W
Feedback Vector on Curators
Denoising Autoencoder
CalculateReconstruction
Loss
During Training
Use Denoising Autoencoder (DAE) to uncover the latent representation of user preference on curators.
Latent Representation
![Page 15: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/15.jpg)
Uncover the PreferencesUse Denoising Autoencoder (DAE) to uncover the latent representation of user preference on curators.
Follow
0100
10
… hCV
0100
10
…
W
Feedback Vector on Curators
Denoising Autoencoder
The Reconstructed
Vector
During Prediction
Preference Scores on Curators
Latent Representation
![Page 16: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/16.jpg)
Uncover the PreferencesWe can enrich the preference on curators with preference on items.
TagRead
Highlight
TagRead
Highlight
Follow0100
10
… hC
1001
01
…
V
0100
10
…W
Feedback Vector on Items
Feedback Vector on Curators
![Page 17: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/17.jpg)
Uncover the PreferencesWe can enrich the preference on curators with preference on items.
TagRead
Highlight
TagRead
Highlight
Follow0100
10
… hC
1001
01
…
V
0100
10
…W
Feedback Vector on Items
Feedback Vector on Curators
hI
VI
![Page 18: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/18.jpg)
Uncover the PreferencesA Joint Curator-Item DAE model
+
TagRead
Highlight
TagRead
Highlight
Follow0100
10
… hC
1001
01
… hI
V
VI
0100
10
…
h
1001
01
…
WI
W
Feedback Vector on Items
Shared Latent Factors
Feedback Vector on Curators
![Page 19: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/19.jpg)
What’s Next?The element at the same dimension in and may not correspond to the same latent factor.
TagRead
Highlight
TagRead
Highlight
Follow0100
10
… hC
1001
01
… hI
V
VI
0100
10
…
h
1001
01
…
WI
W
Feedback Vector on Items
Shared Latent Factors
Feedback Vector on Curators
hC hI
![Page 20: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/20.jpg)
What’s Next?How to assign personalized weights on and ?hC hI
TagRead
Highlight
TagRead
Highlight
Follow0100
10
… hC
1001
01
… hI
V
VI
0100
10
…
h
1001
01
…
WI
W
Feedback Vector on Items
Shared Latent Factors
Feedback Vector on Curators
![Page 21: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/21.jpg)
Fusing Latent RepresentationsUse a Discriminator to force and to live in a shared space.
hC hI
TagRead
Highlight
TagRead
Highlight
Follow0100
10
…
hC
1001
01
…
hI
V
VI
Input…
Fully-Connected Layers
Feedback Vector on Items
Adversarial loss for distinguishing
and hC hI
Discriminator
Feedback Vector on Curators
![Page 22: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/22.jpg)
Generate the user-dependent weights for and via an attention layer.
Personalized Fusing
0100
10
…
hC
1001
01
…
hI
V
Vh
h↵C
↵I
VI
0100
10
…
h1001
01
…WI
W
EAttentionLayer
Isolated Latent Factors
Feedback Vector on Items
Shared Latent Factors
Feedback Vector on Curators
hC hI
Input
![Page 23: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/23.jpg)
Generate the user-dependent weights for and via an attention layer.
Personalized Fusing
0100
10
…
hC
1001
01
…
hI
V
Vh
h↵C
↵I
VI
0100
10
…
h1001
01
…WI
W
EAttentionLayer
Isolated Latent Factors
Feedback Vector on Items
Shared Latent Factors
Feedback Vector on Curators
hC hI
Output
![Page 24: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/24.jpg)
CuRe - Curator Recommendation
TagRead
Highlight
TagRead
Highlight
Follow0100
10
…
hC
1001
01
…
hI
V
Vh
h↵C
↵I
VI
0100
10
…
h1001
01
…WI
W
EAttentionLayer
Input
Isolated Latent Factors
…
Fully-Connected Layers
Feedback Vector on Items
Adversarial loss for distinguishing
and hC hI
Discriminator
Shared Latent Factors
Feedback Vector on Curators
![Page 25: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/25.jpg)
Experiment: Data
Dataset #User #Item #User-User Interactions
#User-Item Interactions
Goodreads 48,208 61,848 528,816 10,526,215
Spotify 25,471 70,107 227,024 4,499,741
Two Datasets:
![Page 26: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/26.jpg)
Experiment: Metric• F1@K: combination of recall and precision
• NDCG@K: takes the position of recommendations into consideration
• K=5, 10
![Page 27: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/27.jpg)
Experiment: BaselinesCompare with the widely used recommendation frameworks:
• MP: Most Popular
• UCF: User-based collaborative filtering
• BPR: Matrix Factorization with Bayesian Personalized Ranking
![Page 28: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/28.jpg)
Experiment: BaselinesCompare with recommendation frameworks enhanced with an adversarial component or built on Autoencoder:
• AMF: Adversarial Matrix Factorization
• DAE: Denoising Autoencoder
• CDAE: Collaborative Denoising Autoencoder
• VAE: Variational Autoencoder for Collaborative Filtering
![Page 29: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/29.jpg)
Experiment: BaselinesAdditional Approaches considering both user-user and user-item interactions:
• EMJ: Embedding Factorization odes for Joint Recommendation
• Joint-DAE: A simplified version of CuRe without adversarial learning process and the attention layer.
![Page 30: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/30.jpg)
CuRe vs Baselines• The proposed model outperforms the state-of-the-art in
recommending curators (by 18% in Goodreads, 6% in Spotify).
• Simultaneously, it is able to achieve significant improvements in item recommendation compared with the baselines.
• Larger improvements under the cold-start setting.
![Page 31: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/31.jpg)
Impact of each component?Utilizing feedback on items can help in inferring preferences on curators.
DAE
Joint-DAEIncorporate the
preference on items
![Page 32: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/32.jpg)
Impact of each component?The adversarial component enables the model to achieve better performance in less epochs.
Adversarial Joint-DAE
Add the Discriminator into the training
process
![Page 33: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/33.jpg)
Impact of each component?Providing personalized fusing is important for achieving the improved performance in both tasks.
Adversarial Joint-DAE + Attention Layers
With personalized fusing layer
![Page 34: Session 01-Jianling Wang-User Recommendation in Content ...people.tamu.edu/~jwang713/slides/curator_wsdm2020.pdf · Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee](https://reader036.vdocuments.site/reader036/viewer/2022081616/600311fd23630f405b4a5c19/html5/thumbnails/34.jpg)
Conclusion• New Problem - Curator Recommendation
• Joint Recommendation for a primary and a supplementary task.
• Experiments prove that the proposed models can outperform the state-of-the-art in both the primary and the supplementary tasks.
• The next step…
• Can we support various types of interactions between users?
• How to capture the temporally dynamic patterns of curators?