Collaborative Filtering: A Machine Learning Perspective

By Benjamin M. Marlin. 2004

223 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Variational Autoencoders for Collaborative Filtering (2018, neutral): “ to [0, 1] after dividing by the best possible DCG@R, where all the held-out items are ranked at the top. 4.3 Experimental setup We study the performance of various models under strong generalization [29]: We split all users into training/validation/test sets. We train models using the entire click ”
Deep Exponential Families (2014, positive): “We re-use the testing procedure introduced in the previous section (this is referred to as strong-generalization in the recommendation literature [22]) where the held-out test set contains one thousand users.”
Improving regularized singular value decomposition for collaborative filtering (2007, unknown): “Another approach is using graphical models [7, 8].”
Controllable Multi-Interest Framework for Recommendation (2020, unknown): “We evaluate the performance of all methods under strong generalization [34, 37, 38]: We split all users into training/validation/test sets by the proportion of 8:1:1.”
Matchbox: large scale online bayesian recommendations (2009, unknown): “In order to measure accuracy in a cold start situation we adopt the methodology of [11] and [13].”
A correlated topic model of Science (2007, unknown): “The correlated topic model builds on the earlier latent Dirichlet allocation (LDA) model of Blei, Ng and Jordan [8], which is an instance of a general family of mixed membership models for decomposing data into multiple latent components.”
Comparison of collaborative filtering algorithms (2011, unknown): “In the literature there are different approaches, most related to machine learning [Marlin 2004]: based on linear algebra methods (SVD [Billsus and Pazzani 1998; Sarwar et al. 2000], factor analysis [Canny 2002], PCA [Goldberg et al. 2001], MMMF [Rennie and Srebro 2005], etc.), clustering [Ungar…”
Ranking with non-random missing ratings: influence of popularity and positivity on evaluation metrics (2012, unknown): “We apply the models in the setting of strong generalization [7], to avoid making too many comparisons but still be in the most difficult evaluation setting.”
Learning from Incomplete Ratings Using Non-negative Matrix Factorization (2006, unknown): “Marlin’s thesis [12] compares a number of algorithms from these two categories for accuracy on available data sets.”
Non-linear matrix factorization with Gaussian processes (2009, unknown): “Our approach outperforms the URP and Attitude algorithms (Marlin, 2004a), the MMMF (Rennie & Srebro, 2005), E-MMMF of (DeCoste, 2006), and the Item-based approach of (Park & Pennock, 2007) 2004a) users with fewer than 20 ratings were discarded; This left us with 36 , 656 users.”
Collaborative Filtering and the Missing at Random Assumption (2007, unknown): “Collaborative (cid:12)ltering research within the machine learning community has focused almost exclusively on developing new models and new learning procedures to improve rating prediction performance [2, 4, 5, 6, 8].”
Addressing cold-start problem in recommendation systems (2008, unknown): “We give out some results of probabilistic models implemented by Marlin [1].”
The multiple multiplicative factor model for collaborative filtering (2004, unknown): “We apply a strong generalization experimental protocol (Marlin, 2004, p. 14).”
Collaborative prediction using ensembles of Maximum Margin Matrix Factorizations (2006, unknown): “As reported in (Rennie & Srebro, 2005) (and summarized in Tables 1 and 3 later), the prediction performance of MMMF on large benchmark rating prediction tasks are already impressive, beating the previous best reported in (Marlin, 2004) by signi(cid:12)cant amounts.”
A Survey of Collaborative Filtering Techniques (2009, unknown): “A multinomial mixture model assumes that there are multiple types of users underlying all profiles, and that the rating variables are independent with each other and with the user’s identity given the user’s type [98].”
Correlated Topic Models (2005, unknown): “Recent work has used LDA as a module in more complicated document models [8, 10, 6], and in a variety of settings such as image processing [11], collaborative ﬁltering [7], disability survey data [4], population genetics [9], and the modeling of sequential data and user proﬁles [5].”
Fast maximum margin matrix factorization for collaborative prediction (2005, unknown): “We mimic the setup used by Marlin (2004) and compare against his results.”