Learning Attribute-to-Feature Mappings for Cold-Start Recommendations

By Zeno Gantner, Lucas Drumond, Christoph Freudenthaler, Steffen Rendle, Lars Schmidt-Thieme. 2010

322 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Ask the GRU: Multi-task Learning for Deep Text Recommendations (2016, methodology): “[30] incorporate side information through a linear regression based formulation on metadata like category, user’s age, location, etc.”
Sound and Music Recommendation with Knowledge Graphs (2016, methodology): “BPR Linear is a hybrid matrix factorization method able to work with sparse datasets [Gantner et al. 2010].”
Collaborative Filtering beyond the User-Item Matrix (2014, background): “One of the most common side information sources is attribute information [Agarwal and Chen 2009; Bao et al. 2009; Gantner et al. 2010a; Koenigstein et al. 2011; Li et al. 2010; Moshfeghi et al. 2011; Shi et .”
Coupled Matrix Factorization Within Non-IID Context (2014, background): “For example, many researchers try to influence or precisely estimate the latent factors [7] [19] [1] [13] through considering the attributes or topics information of users and items for latent factor models.”
Recommendations : Modeling Music Ratings with Temporal Dynamics and Item Taxonomy (2011, unknown): “some of the more recent approaches [1, 8, 9].”
Multi-relational matrix factorization using bayesian personalized ranking for social network data (2012, background): “BPR-map [1] also is a two-step process.”
Heterogeneous Information Network Embedding for Recommendation (2017, background): “[22] explain drawbacks of three popular cold-start models [23], [24], [25] and further propose a learning based approach for the cold-start problem to leverage social data via randomised SVD.”
A Survey on Knowledge Graph-Based Recommender Systems (2020, background): “Common options for user side information involve user’s demographic information [63], including occupation, gender, and hobbies; and user network [11], [12].”
Relational Collaborative Topic Regression for Recommender Systems (2015, methodology): “Hybrid methods [2], [3], [6], [12], [15], [16], [17], [20], [32], [34], [40], [43], [47], [48] combine both content-based methods and CF based methods by ensemble techniques.”
CityTransfer (2018, methodology): “Gantner et al. [12] describe a method that maps entity (e.g. new user or item) attributes to the latent features of a matrix factorization model.”
Improving the quality of predictions using textual information in online user reviews (2013, background): “In [11], the authors enhance a matrix factorization-based recommendation system by mapping user or item attributes to the latent factors to make predictions for new users or new items.”
Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5) (2022, background): “Solutions to this problem either learn to model content features [15, 29, 44, 55] so that inference can be made without interaction records or learn to transfer representations from auxiliary domains [42, 56, 59, 72, 82].”
Sparse linear methods with side information for top-n recommendations (2012, methodology): “Such approaches include hybrid methods [5], matrix/tensor factorization [14, 8], and other regression methods [1].”
LARA: Attribute-to-feature Adversarial Learning for New-item Recommendation (2020, methodology): “• BPR-kNN [12]: This is a two-step method for cold-start recommendation.”
Link Prediction via Matrix Factorization (2011, background): “Recent advances in this literature are based on inferring reasonable values of latent features by falling back to the side information as a prior [2,12].”
A Generic Coordinate Descent Framework for Learning from Implicit Feedback (2016, methodology): “Feature-based factorization models are strictly more powerful than MF and have shown large improvements in many applications (e.g. [2, 11]).”