User Recommendation in Content Curation Platforms Jianling Wang, Ziwei Zhu and James Caverlee
Content Creation vs Curation Content creators generate new digital artifacts such as tweets, blog posts, or photos.
Content Creation vs Curation Music Streaming platforms allow users to create and share playlists. Playlists Mega Hit Mix Jogging! Mood Booster
Content Creation vs Curation Goodreads provide a platform for users to curate interesting books via tagging, ratings and reviews. John Smith Following John added to want to read John rated it John added to Bio Preview Book Like Comment Preview Book Preview Book Like Comment Like Comment
Our Goal: Recommend Curators 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 Compared 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.
Why Recommend Curators? • 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 di ff erent genres). Playlists For study Jogging! Mood Booster
Why Recommend Curators? • By receiving updates from whom they follow, users can be exposed to interesting items and curation decisions. Item g T a n i s t e L t e R a G e t o w F o l l e d a t U p User
Our Setting We can collect: Feedback Vector • User-curator following relationships on Curators 0 1 Follow • Implicit feedback on items 0 0 … 1 Tag 0 Read Highlight 1 0 0 Tag 1 Read Highlight … 0 1 Feedback Vector on Items
Challenge How to model these two aspects - curator preferences and item preferences - in a unified model? Feedback Vector on Curators 0 1 Follow 0 0 … 1 Tag 0 Read Highlight 1 0 0 Tag 1 Read Highlight … 0 1 Feedback Vector on Items
The Goal We are motivated to develop a new model for Cu rator Re commendation that leverages the linkage between user- curator following relationships and the items they are interested in.
The Joint Tasks Ultimately, the model aims to provide users with recommendation on: • who to follow (the primary task) • interesting items (the supplementary task)
CuRe - Cu rator Re commendation Three components: • Learning Curator & Item Preferences • Fusing Latent Representations • Personalized via Attention
Uncover the Preferences Use Denoising Autoencoder (DAE) to uncover the latent representation of user preference on curators. Feedback Vector on During Curators Training 0 0 1 1 0 0 Calculate Follow V W h C 0 0 Reconstruction … … Latent Loss 1 1 Representation 0 0 Denoising Autoencoder
Uncover the Preferences Use Denoising Autoencoder (DAE) to uncover the latent representation of user preference on curators. Feedback Vector on During Curators Prediction 0 0 1 1 0 0 The Follow V W h C 0 0 Reconstructed … … Latent Vector 1 1 Representation 0 0 Preference Denoising Scores on Autoencoder Curators
Uncover the Preferences We can enrich the preference on curators with preference on items. Feedback Vector on Curators 0 0 1 1 Follow 0 0 V W h C 0 0 … … 1 1 Tag 0 0 Read Highlight 1 0 Tag 0 Read 1 Highlight … 0 1 Feedback Vector on Items
Uncover the Preferences We can enrich the preference on curators with preference on items. Feedback Vector on Curators 0 0 1 1 Follow 0 0 V W h C 0 0 … … 1 1 Tag 0 0 Read Highlight 1 0 Tag 0 Read 1 h I Highlight V I … 0 1 Feedback Vector on Items
Uncover the Preferences A Joint Curator-Item DAE model Feedback Vector on Curators 0 0 1 1 Follow 0 0 Shared V h C 0 0 Latent … … Factors 1 1 Tag W 0 0 + Read h Highlight 1 1 W I 0 0 0 Tag 0 1 h I 1 Read V I Highlight … … 0 0 1 1 Feedback Vector on Items
What’s Next? The element at the same dimension in and may not h C h I correspond to the same latent factor. Feedback Vector on Curators 0 0 1 1 Follow 0 0 Shared V h C 0 0 Latent … … Factors 1 1 Tag W 0 0 Read h Highlight 1 1 W I 0 0 0 Tag 0 1 h I 1 Read V I Highlight … … 0 0 1 1 Feedback Vector on Items
What’s Next? How to assign personalized weights on and ? h C h I Feedback Vector on Curators 0 0 1 1 Follow 0 0 Shared V h C 0 0 Latent … … Factors 1 1 Tag W 0 0 Read h Highlight 1 1 W I 0 0 0 Tag 0 1 h I 1 Read V I Highlight … … 0 0 1 1 Feedback Vector on Items
Fusing Latent Representations Use a Discriminator to force and to live in a shared h I h C space. Feedback Vector on Curators 0 1 Follow 0 0 Fully-Connected Layers … V h C 1 Tag Adversarial loss for 0 Read … distinguishing Input Highlight and h C h I 1 0 h I 0 Tag V I 1 Read Discriminator … Highlight 0 1 Feedback Vector on Items
Personalized Fusing Generate the user-dependent weights for and via an h I h C attention layer. Feedback Vector Isolated on Curators Latent 0 0 Factors 1 1 V 0 0 h 0 0 W Shared … V … Latent α C h C h 1 1 Factors 0 0 h 1 1 0 0 h I α I 0 0 V I 1 W I 1 … … 0 0 1 1 Attention E Feedback Vector Layer Input on Items
Personalized Fusing Generate the user-dependent weights for and via an h I h C attention layer. Feedback Vector Isolated on Curators Latent 0 0 Factors 1 1 V 0 0 h 0 0 W Shared … V … Latent α C h C h 1 1 Factors 0 0 h 1 1 0 0 h I α I 0 0 V I 1 W I 1 Output … … 0 0 1 1 Attention E Feedback Vector Layer on Items
CuRe - Cu rator Re commendation Feedback Vector Isolated on Curators Latent 0 0 Factors 1 1 V Follow 0 0 h 0 0 W Shared Fully-Connected Layers … V … Latent h C α C h 1 1 Factors Tag Adversarial loss for 0 0 Read … distinguishing Input h Highlight h C and h I 1 1 0 0 h I α I 0 0 Tag V I W I 1 1 Read Discriminator … … Highlight 0 0 1 1 Attention E Feedback Vector Layer on Items
Experiment: Data Two Datasets: #User-User #User-Item Dataset #User #Item Interactions Interactions Goodreads 48,208 61,848 528,816 10,526,215 Spotify 25,471 70,107 227,024 4,499,741
Experiment: Metric • F1@K: combination of recall and precision • NDCG@K: takes the position of recommendations into consideration • K=5, 10
Experiment: Baselines Compare with the widely used recommendation frameworks: • MP : Most Popular • UCF : User-based collaborative filtering • BPR : Matrix Factorization with Bayesian Personalized Ranking
Experiment: Baselines Compare 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
Experiment: Baselines Additional 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.
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.
Impact of each component? Utilizing feedback on items can help in inferring preferences on curators. Joint-DAE Incorporate the preference on items DAE
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
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
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?
Recommend
More recommend
