Abstract:User-generated item lists are popular on many platforms. Examples include video-based playlists on YouTube, image-based lists (or"boards") on Pinterest, book-based lists on Goodreads, and answer-based lists on question-answer forums like Zhihu. As users create these lists, a common challenge is in identifying what items to curate next. Some lists are organized around particular genres or topics, while others are seemingly incoherent, reflecting individual preferences for what items belong together. Furthermore, this heterogeneity in item consistency may vary from platform to platform, and from sub-community to sub-community. Hence, this paper proposes a generalizable approach for user-generated item list continuation. Complementary to methods that exploit specific content patterns (e.g., as in song-based playlists that rely on audio features), the proposed approach models the consistency of item lists based on human curation patterns, and so can be deployed across a wide range of varying item types (e.g., videos, images, books). A key contribution is in intelligently combining two preference models via a novel consistency-aware gating network - a general user preference model that captures a user's overall interests, and a current preference priority model that captures a user's current (as of the most recent item) interests. In this way, the proposed consistency-aware recommender can dynamically adapt as user preferences evolve. Evaluation over four datasets(of songs, books, and answers) confirms these observations and demonstrates the effectiveness of the proposed model versus state-of-the-art alternatives. Further, all code and data are available at https://github.com/heyunh2015/ListContinuation_WSDM2020.

Consistence beats causality in recommender systems.

Personalized Recommendation Based on Unbiased Consistence

User Interest Dynamics on Personalized Recommendation

Redundant Correlation Effect on Personalized Recommendation

A Personalized Recommendation Algorithm Via Biased Random Walk

Can Dissimilar Users Contribute to Accuracy and Diversity of Personalized Recommendation?

Temporal Information Effect on Personalized Recommendation

Recommendation over Time: a Probabilistic Model of Time-Aware Recommender Systems

Predictability of diffusion-based recommender systems

C$^2$-Rec: An Effective Consistency Constraint for Sequential Recommendation

Consistency-Aware Recommendation for User-Generated ItemList Continuation

Solving the Apparent Diversity-Accuracy Dilemma of Recommender Systems

Interest Before Liking: Two-step Recommendation Approaches

CSRec: Rethinking Sequential Recommendation from A Causal Perspective

Learning the Dynamics in Sequential Recommendation by Exploiting Real-time Information

CauseRec: Counterfactual User Sequence Synthesis for Sequential Recommendation

TaDb: A Time-Aware Diffusion-Based Recommender Algorithm

Personalized Recommender Systems Based on Social Relationships and Historical Behaviors

Causal Collaborative Filtering

A Time-aware Hybrid Algorithm for Online Recommendation Services

Solving the Stability–accuracy–diversity Dilemma of Recommender Systems