Music Recommenders: User Evaluation Without Real Users?

Authors: Susan Craw, Ben Horsburgh, Stewart Massie

IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that its findings are consistent with those from an online study with real users. This paper explores system-centric evaluation based on user data from social media sites relevant to music.
Researcher Affiliation Academia Susan Craw and Ben Horsburgh and Stewart Massie IDEAS Research Institute and School of Computing Science & Digital Media Robert Gordon University, Aberdeen, UK {s.craw, s.massie}@rgu.ac.uk
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes The Million Song Dataset [MSD, nd; Bertin-Mahieux et al., 2011] is a step in the right direction, but provides audio data as pre-extracted features, and so cannot be used to evaluate recommenders based on other audio features. Our music collection is Horsburgh s [2013] dataset containing 3174 audio tracks by 764 separate artists.
Dataset Splits No The paper does not specify explicit training, validation, and test dataset splits with percentages or sample counts for model training. It describes a 'user study' as a 'benchmark to validate' the social Sim evaluation, but this is a conceptual validation, not a data split for model training.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions software like 'Last.fm s API' and 'Vamp plugin' but does not provide specific version numbers for these or any other software dependencies, which would be necessary for reproducible setup.
Experiment Setup Yes A k-NN nearest-neighbour retrieval using cosine similarity in the MFS-LSI space identifies the K = 40 most similar audio tracks. The number of pseudo-tags P from p to be included in the hybrid vector h, to balance the number of existing tags T in t. A weighting α determines the influence of selected pseudo-tags p on h1. h = α p + (1 α)t where P = 100 T α = 0.5 P/100