Gamma-Poisson Dynamic Matrix Factorization Embedded with Metadata Influence

Authors: Trong Dinh Thac Do, Longbing Cao

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that m GDMF significantly (both effectively and efficiently) outperforms the state-of-the-art static and dynamic models on large, sparse and dynamic data.
Researcher Affiliation Academia Trong Dinh Thac Do Advanced Analytics Institute University of Technology Sydney thacdtd@gmail.com Longbing Cao Advanced Analytics Institute University of Technology Sydney longbing.cao@gmail.com
Pseudocode Yes Algorithm 1 SVI for m GDMF
Open Source Code No The paper does not provide any statement or link indicating the availability of open-source code for the methodology.
Open Datasets Yes Netflix-Time. Similar procedure as in [37, 16, 34] is taken to obtain a subset of Netflix Prize data [4]... Yelp-Active. A subset of the Yelp Academic Challenge data is obtained similarly to [34]... LFM-Tracks. It contains the number of times a user listened to a song during a given time period [12]: 16 time slices of 0.9K users and 1M tracks (i.e., songs), similar to [34];
Dataset Splits Yes We then randomly sample and assign 5% of the test set for validation, similar to [16, 34].
Hardware Specification No The paper does not provide specific details about the hardware used for experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers used for the experiments.
Experiment Setup Yes For the static portions, we set a = b = c = d = 0.3 in the same way as in HPF. The metadata hyper-parameters a , b , c and d are set to a small value: 0.1, so that the user/item attribute weights automatically grow over time. We also set aθ = aγ = aθ = bθ = bβ = aι = 1 to keep the chains small at the beginning. We test a wide range of latent components K from 10 to 200 and choose the best K = 50 for m GDMF/GDMF. For SVI hyper-parameters, we assign 10, 000 as the learning rate delay iter0 and 0.7 as the learning rate power ϵ, similar to [34] and [3].