Evaluating Recommender System Stability with Influence-Guided Fuzzing

Authors: David Shriver, Sebastian Elbaum, Matthew B. Dwyer, David S. Rosenblum4934-4942

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We implement our approach and evaluate it on several recommender algorithms using the Movie Lens dataset. We find that influence-guided fuzzing can effectively find small sets of modifications that cause significantly more instability than random approaches.
Researcher Affiliation Academia 1Department of Computer Science, University of Virginia, Charlottesville, VA, USA 2Department of Computer Science, National University of Singapore, Singapore
Pseudocode No The paper defines functions and heuristics mathematically but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper provides a link (https://github.com/lyst/lightfm) for the Light FM algorithm, which is a third-party tool used in their study, not their own source code for the methodology described in the paper.
Open Datasets Yes Movie Lens 100k Dataset We use the dataset released in 1998 from the Movie Lens recommendation system. The dataset is available as a group of tab separated files, containing 100 thousand integer ratings (from 1 to 5) collected from 943 users over a period of 8 months on 1682 movies. Each user has rated at least 20 items. More details about the data collection process and the dataset itself are available at https://grouplens.org/datasets/movielens/100k/ (Harper and Konstan 2015).
Dataset Splits No The paper mentions modifying the Movie Lens 100k dataset and using it for training, but it does not specify any explicit train/validation/test splits of the original dataset for model training or evaluation in the standard sense (e.g., 80/10/10 split).
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions using the 'Lenskit framework' and a 'Python implementation' for Light FM, but it does not provide specific version numbers for these software components or any other libraries, which are necessary for reproducible dependency information.
Experiment Setup Yes We evaluated the fuzzing heuristics for 3 sizes of modification set: 1, 10, and 100. [...] For each heuristic, size, and recommender system (described next), we generated 100 modification sets. We then trained each recommender on the modified dataset and generate Top-10 recommendations for every user. [...] Funk SVD ... learns 25 latent features. [...] We experimented with several values of ϵ, and we chose 0.05 as the value that generally produced the most instability.