SAGA: A Submodular Greedy Algorithm for Group Recommendation

Authors: Shameem Puthiya Parambath, Nishant Vijayakumar, Sanjay Chawla

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we describe the experimental setup and report the results using user groups generated from the 1M Movie Lens1 dataset.
Researcher Affiliation Collaboration Shameem A Puthiya Parambath QCRI, HBKU, Doha, Qatar spparambath@hbku.edu.qa Nishant Vijayakumar Apptopia Inc., Boston, USA nishant.vijayakumar@gmail.com Sanjay Chawla QCRI, HBKU, Doha, Qatar schawla@hbku.edu.qa
Pseudocode Yes Algorithm 1: SAGA: Submodular Greedy Group Recommendation Algorithm
Open Source Code No The paper does not provide any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes In this section, we describe the experimental setup and report the results using user groups generated from the 1M Movie Lens1 dataset. 1http://grouplens.org/datasets/movielens/
Dataset Splits No We carried out holdout validation by randomly selecting 30% of the item set and marking it as unrated wherever the rating values are observed.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions using a 'non-negative matrix factorization based on weighted-regularized non-negative alternating least squares algorithm' but does not specify any software names with version numbers (e.g., programming languages, libraries, or solvers).
Experiment Setup Yes In our experiments, the dimension of the item/user feature space is set to 150. We used the rbf kernel as the item afinity function h, i.e. Wij = exp( γ||xi xj||2) where xi and xj are the ith and jth item features. The γ value is chosen by running the algorithm for a set of values in the range {2 3, , 23} in multiples of two, and the reported results are for the best γ value for the respective algorithms. For the item saturation function f, we use natural logarithm f(x) = ln(1 + x), and for the user saturation function we experimented with two settings (i) identity function gu(x) = x and (ii) concave function gu(x) = x.