Probabilistic Matrix Factorization with Non-random Missing Data

Authors: Jose Miguel Hernandez-Lobato, Neil Houlsby, Zoubin Ghahramani

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We analyze the performance of our Matrix Factorization model with data Missing Not At Random (MF-MNAR) in a series of experiments with synthetic and real-world rating data.
Researcher Affiliation Academia Jos e Miguel Hern andez-Lobato JMH233@CAM.AC.UK Neil Houlsby NMTH2@CAM.AC.UK Zoubin Ghahramani ZOUBIN@ENG.CAM.AC.UK University of Cambridge, Department of Engineering, Cambridge CB2 1PZ, UK
Pseudocode Yes Algorithm 1 Approximate Inference in the Joint Model
Open Source Code Yes The code for our MF-MNAR method is publicly available at http://jmhl.org.
Open Datasets Yes The Movie Lens 100k and 1M datasets1 include ratings from 1 to 5 on movies. The Yahoo! Web-scope R3 dataset2. contains ratings from 1 to 5 on songs. We also analyzed a dataset obtained from the reviewer bidding process for the 2013 NIPS conference. Finally, the Movie Tweetings dataset3 includes ratings on movies collected from Twitter.
Dataset Splits No we randomly split the observed ratings (the ri,j with xi,j = 1) into a training set with 99% of the ratings and a standard test set with the remaining 1%. (No mention of a dedicated validation set.)
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments (e.g., CPU/GPU models, memory).
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes In all MF methods we use a latent rank of size 20. In the mixture of multinomials we use 20 components.